feat(monitor): 抖音监控改走新的 Web 接口客户端（接上上一版的移植）

上一版只把客户端写出来、验证了它单独可用，**但没有接进任何地方** —— 所以你建的任务跑起来
仍然在调爬虫子进程，报的仍然是那句自己编的「account blocked」。这一步把它接上。

* runner 的 Phase 2 按平台分岔：dy 走进程内 HTTP 客户端（douyin_fetch），其余平台照旧走
  爬虫子进程。抖音那条不再起 Playwright、不再构造那串自相矛盾的浏览器指纹参数。
* 新增 douyin_fetch：把采到的东西写成 store/douyin 那套 jsonl 形状 —— **ingest 完全不知道
  数据是从哪来的**，重采样/差分/事件/通知/报表全都照旧，一个字没改。
* 失败不再假装：一条都没采到就以非零退出码 + **真实原因**交给 ingest，落成
  「采集进程异常退出（code=1）：…」。绝不会再掉进「疑似登录失效」那个分支。
* 已知作品列表接口（aweme/post）被抖音单独加了真校验（200 + 空 body），所以加了退化：
  拿不到列表就用库里已知的 aweme_id 逐条走 detail 刷新。**边界是：已知作品的指标能继续
  更新，新作品发现不了** —— 这个边界会以一条 warning 日志留下痕迹，不让它看起来一切正常。
* 顺带给客户端补上 video_detail（实测可用：200 / 45425 字节），退化路径靠它。

测试 +6：产物目录与文件名、评论文件即使为空也要建（ingest 靠它区分「没评论」和
「什么都没抓到」）、重复作品只写一次、列表被挡时的退化、彻底失败仍写出产物与原因、
评论失败不连累作品。

api/monitor/runner.py

@@ -28,6 +28,8 @@ import os
 from pathlib import Path
 from typing import Iterable, List, Optional
 
+from sqlalchemy import select
+
 from tools.time_util import get_current_timestamp
 
 from ..schemas import (
@@ -38,7 +40,7 @@ from ..schemas import (
     SaveDataOptionEnum,
 )
 from ..services import crawler_manager
-from . import adapters, app_settings, covers, notify
+from . import adapters, app_settings, covers, douyin_fetch, notify
 from .db import get_session
 from .ingest import IngestResult, diagnose_failure, ingest_run
 from .models import (
@@ -47,6 +49,7 @@ from .models import (
     RUN_PENDING,
     RUN_RUNNING,
     RUN_TIMEOUT,
+    MonitorNote,
     MonitorRun,
     MonitorTarget,
     MonitorTask,
@@ -194,57 +197,102 @@ async def execute_task(task_id: int, trigger: str = "manual") -> IngestResult:
         max_notes_count = task.max_notes_count
         max_comments_count = task.max_comments_count
         timeout_seconds = task.run_timeout_seconds
+        # 抖音的作品列表接口被那道真校验挡着（见 douyin_fetch），拿不到列表时就靠这些
+        # 已知的 aweme_id 逐条刷新 —— 新作品发现不了，但已有作品的指标还能继续更新。
+        known_aweme_ids = list(
+            await session.scalars(
+                select(MonitorNote.note_id).where(MonitorNote.task_id == task.id)
+            )
+        )
 
-    # --- Phase 2: run the crawler outside any transaction ---------------------
-    cookie_file = out_dir / ".cookies"
-    _write_cookie_file(cookie_file, cookie)
+    # --- Phase 2: run the collection outside any transaction ------------------
+    # 抖音走**进程内 HTTP 客户端**，不起 Playwright 子进程：那边会构造一大串自相矛盾的
+    # 浏览器指纹参数（参数说 Mac + Chrome 125、UA 说 Linux + Chrome 155），网关回一个
+    # 200 + 空 body，然后被翻译成「account blocked」—— 看着像账号被封，其实什么都不是。
+    # 见 douyin_fetch / douyin_api。
+    in_process_tail: List[str] = []
+    if platform == adapters.PLATFORM_DY:
+        fetched = await douyin_fetch.collect(
+            out_dir,
+            platform=platform,
+            mode=mode,
+            limit=max_notes_count,
+            want_comments=enable_comments,
+            comment_limit=max_comments_count,
+            targets=targets,
+            known_aweme_ids=known_aweme_ids,
+            cookie=cookie,
+        )
+        in_process_tail = list(fetched["errors"])
+        # 一条都没采到 = 这一轮失败，并把**真因**当作退出诊断传下去。否则它会掉进
+        # ingest 的「疑似登录失效」分支 —— 又骗人一次，正是这套东西一直在犯的毛病。
+        exit_code = 1 if (fetched["errors"] and not fetched["notes"]) else 0
+        if fetched["errors"] and fetched["notes"]:
+            # 有产物但带着错误，说明走了退化路径（比如作品列表被挡，只刷新了已知作品）。
+            # 这一轮状态是成功，但**不是**一切正常 —— 得留下痕迹，否则没人知道新作品
+            # 其实没在发现。
+            print(
+                "[monitor.runner] 抖音采集部分失败："
+                + "；".join(fetched["errors"])[:300]
+            )
+    else:
+        cookie_file = out_dir / ".cookies"
+        _write_cookie_file(cookie_file, cookie)
 
-    request = CrawlerStartRequest(
-        platform=_PLATFORM_ENUM[platform],
-        login_type=LoginTypeEnum.COOKIE,
-        crawler_type=CrawlerTypeEnum.CREATOR if mode == MODE_CREATOR else CrawlerTypeEnum.DETAIL,
-        creator_ids=",".join(urls) if mode == MODE_CREATOR else "",
-        specified_ids=",".join(urls) if mode != MODE_CREATOR else "",
-        start_page=1,
-        enable_comments=enable_comments,
-        enable_sub_comments=strategy["enable_sub_comments"],
-        enable_media=False,
-        save_option=SaveDataOptionEnum.JSONL,
-        cookies="",
-        headless=True,
-        max_notes_count=max_notes_count,
-        max_comments_count=max_comments_count,
-        # Isolate this run's output: the crawler names files by date only, so
-        # otherwise same-day runs would append into one shared file.
-        save_data_path=str(out_dir),
-        # Attach to the browser already running on CDP_DEBUG_PORT when the
-        # operator enabled it; otherwise launch a private, throwaway browser.
-        enable_cdp_mode=cdp_enabled,
-        # Only injecting web_session is not enough to sign requests from a cold
-        # browser profile.
-        inject_all_cookies=True,
-        save_login_state=True,
-        cookies_file=str(cookie_file),
-        max_concurrency_num=1,
-        # Strategy + proxy, surfaced on the Settings page.
-        crawler_max_sleep_sec=strategy["crawl_sleep_sec"],
-        enable_ip_proxy=strategy["enable_ip_proxy"],
-        ip_proxy_pool_count=strategy["proxy_pool_count"],
-        ip_proxy_provider_name=strategy["proxy_provider"],
-        static_proxy_url=strategy["static_proxy_url"] or None,
+        request = CrawlerStartRequest(
+            platform=_PLATFORM_ENUM[platform],
+            login_type=LoginTypeEnum.COOKIE,
+            crawler_type=CrawlerTypeEnum.CREATOR if mode == MODE_CREATOR else CrawlerTypeEnum.DETAIL,
+            creator_ids=",".join(urls) if mode == MODE_CREATOR else "",
+            specified_ids=",".join(urls) if mode != MODE_CREATOR else "",
+            start_page=1,
+            enable_comments=enable_comments,
+            enable_sub_comments=strategy["enable_sub_comments"],
+            enable_media=False,
+            save_option=SaveDataOptionEnum.JSONL,
+            cookies="",
+            headless=True,
+            max_notes_count=max_notes_count,
+            max_comments_count=max_comments_count,
+            # Isolate this run's output: the crawler names files by date only, so
+            # otherwise same-day runs would append into one shared file.
+            save_data_path=str(out_dir),
+            # Attach to the browser already running on CDP_DEBUG_PORT when the
+            # operator enabled it; otherwise launch a private, throwaway browser.
+            enable_cdp_mode=cdp_enabled,
+            # Only injecting web_session is not enough to sign requests from a cold
+            # browser profile.
+            inject_all_cookies=True,
+            save_login_state=True,
+            cookies_file=str(cookie_file),
+            max_concurrency_num=1,
+            # Strategy + proxy, surfaced on the Settings page.
+            crawler_max_sleep_sec=strategy["crawl_sleep_sec"],
+            enable_ip_proxy=strategy["enable_ip_proxy"],
+            ip_proxy_pool_count=strategy["proxy_pool_count"],
+            ip_proxy_provider_name=strategy["proxy_provider"],
+            static_proxy_url=strategy["static_proxy_url"] or None,
+        )
+
+        async with get_session() as session:
+            run = await session.get(MonitorRun, run_id)
+            if run is not None:
+                run.status = RUN_RUNNING
+                run.started_at = get_current_timestamp()
+
+        try:
+            exit_code = await crawler_manager.run_and_wait(request, timeout=timeout_seconds)
+        finally:
+            _remove_cookie_file(cookie_file)
+
+    # 失败时的诊断来源。抖音那条路没有子进程，尾巴就是它自己报的错 —— **别去读
+    # crawler_manager 的尾巴**，那里面是上一轮别的平台留下的东西，会张冠李戴。
+    output_tail = (
+        in_process_tail
+        if platform == adapters.PLATFORM_DY
+        else crawler_manager.get_output_tail()
     )
 
-    async with get_session() as session:
-        run = await session.get(MonitorRun, run_id)
-        if run is not None:
-            run.status = RUN_RUNNING
-            run.started_at = get_current_timestamp()
-
-    try:
-        exit_code = await crawler_manager.run_and_wait(request, timeout=timeout_seconds)
-    finally:
-        _remove_cookie_file(cookie_file)
-
     # --- Phase 3: ingest ------------------------------------------------------
     async with get_session() as session:
         run = await session.get(MonitorRun, run_id)
@@ -261,7 +309,7 @@ async def execute_task(task_id: int, trigger: str = "manual") -> IngestResult:
             run.error_message = "Run was killed by timeout or failed to start"
             # -1 同时代表「超时」和「根本没起来」，两者要查的东西完全不同。输出尾巴里
             # 有异常就带上它，否则运行历史里只能看到这句没有信息量的话。
-            cause = diagnose_failure(crawler_manager.get_output_tail())
+            cause = diagnose_failure(output_tail)
             if cause:
                 run.error_message = f"{run.error_message}；原因：{cause}"
             result = IngestResult(status=RUN_TIMEOUT, error=run.error_message)
@@ -271,7 +319,7 @@ async def execute_task(task_id: int, trigger: str = "manual") -> IngestResult:
             # 把爬虫输出的尾巴交给 ingest：退出码本身说明不了问题，运行历史里要显示的
             # 是真正的报错（比如抖音的 DataFetchError: account blocked）。
             result = await ingest_run(
-                session, run, task, out_dir, output_tail=crawler_manager.get_output_tail()
+                session, run, task, out_dir, output_tail=output_tail
             )
 
         # A run that authenticated fine is the only useful signal that the