fix(backend): 部署韧性——模型自愈/就绪门禁/全局代理/启动诊断

- whisper: model.bin 截断/损坏时删目录重下重试一次，修「Unable to
  open file model.bin」死循环；mlx 同样按 config.json 判完整性
- /generate_note 加就绪门禁：本地转写引擎模型没下好直接拦截，返回
  reason=transcriber_model_not_ready，不让任务静默卡在首次下载
- 全局代理：新增 ProxyConfigManager（JSON 配置 + HTTP_PROXY env 兜底）
  + build_openai_client，统一注入代理到 LLM/Groq 客户端；yt-dlp 与
  youtube-transcript-api 也走代理
- build_openai_client 校验 api_key 非空，空 key 给「xxx 的 API Key
  未配置」而不是天书般的 Illegal header value b'Bearer '
- universal_gpt: 模型拒绝自定义 temperature（o1/o3/gpt-5 系列）时
  就地去掉参数重试，不消耗重试预算
- connect_test 改用真实 chat completion 而非 /v1/models 探测
- main.py: lifespan 拆 [startup 1/5..5/5] 分段日志 + 异常清晰定位
- /sys_health 重构为结构化返回 {backend,ffmpeg,db,whisper_model}

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend/app/downloaders/youtube_downloader.py

@@ -9,12 +9,22 @@ from app.downloaders.base import Downloader, DownloadQuality
 from app.downloaders.youtube_subtitle import YouTubeSubtitleFetcher
 from app.models.notes_model import AudioDownloadResult
 from app.models.transcriber_model import TranscriptResult
+from app.services.proxy_config_manager import ProxyConfigManager
 from app.utils.path_helper import get_data_dir
 from app.utils.url_parser import extract_video_id
 
 logger = logging.getLogger(__name__)
 
 
+def _apply_proxy(ydl_opts: dict) -> dict:
+    """YouTube 在国内需要代理。配置了全局代理就塞进 yt-dlp opts。"""
+    proxy = ProxyConfigManager().get_proxy_url()
+    if proxy:
+        ydl_opts['proxy'] = proxy
+        logger.info(f"yt-dlp 走代理: {proxy}")
+    return ydl_opts
+
+
 class YoutubeDownloader(Downloader, ABC):
     def __init__(self):
 
@@ -46,6 +56,7 @@ class YoutubeDownloader(Downloader, ABC):
         if skip_download:
             ydl_opts['skip_download'] = True
 
+        _apply_proxy(ydl_opts)
         with yt_dlp.YoutubeDL(ydl_opts) as ydl:
             info = ydl.extract_info(video_url, download=not skip_download)
             video_id = info.get("id")
@@ -91,6 +102,7 @@ class YoutubeDownloader(Downloader, ABC):
             'merge_output_format': 'mp4',  # 确保合并成 mp4
         }
 
+        _apply_proxy(ydl_opts)
         with yt_dlp.YoutubeDL(ydl_opts) as ydl:
             info = ydl.extract_info(video_url, download=True)
             video_id = info.get("id")

backend/app/downloaders/youtube_subtitle.py

@@ -8,6 +8,7 @@ from typing import Optional, List
 from youtube_transcript_api import YouTubeTranscriptApi
 
 from app.models.transcriber_model import TranscriptResult, TranscriptSegment
+from app.services.proxy_config_manager import ProxyConfigManager
 from app.utils.logger import get_logger
 
 logger = get_logger(__name__)
@@ -17,7 +18,21 @@ class YouTubeSubtitleFetcher:
     """通过 youtube-transcript-api 获取 YouTube 字幕。"""
 
     def __init__(self):
-        self._api = YouTubeTranscriptApi()
+        # 配了全局代理就给 youtube-transcript-api 套一个带 proxies 的 requests.Session，
+        # 否则国内拉字幕同样会超时。代理未配置时退回默认无代理客户端。
+        proxy = ProxyConfigManager().get_proxy_url()
+        if proxy:
+            try:
+                import requests
+                session = requests.Session()
+                session.proxies = {"http": proxy, "https": proxy}
+                self._api = YouTubeTranscriptApi(http_client=session)
+                logger.info(f"YouTube 字幕走代理: {proxy}")
+            except Exception as e:
+                logger.warning(f"为 youtube-transcript-api 注入代理失败，回退无代理: {e}")
+                self._api = YouTubeTranscriptApi()
+        else:
+            self._api = YouTubeTranscriptApi()
 
     def fetch_subtitles(
         self,