feat: 调度器+并发管线+GPU优化+日志分级+前端修复

- scheduler: ffmpeg 异步线程 + GPU 串行调度 + 模型复用（2N→2 次加载）
- pipeline: 阶段拆分（extract/asr/translate），中间数据存 Task 字段
- translate_service: 长度排序批处理，padding 浪费减少 91%
- model_manager: ASR/翻译不共驻，BatchedInferencePipeline 批量解码
- 日志分级: INFO=任务流转里程碑，DEBUG=进度详情；默认 INFO
- 前端: 日志最新在上+滚动感知+退避轮询；24h 时间；上传中状态显示
- /health: 返回完整 Whisper/NLLB 配置
- upload_service: 单事务 complete + 扩展名白名单
- task_router: 合并 UploadSession 虚拟任务到列表
- Dockerfile: CPU/GPU 独立构建链，deps 缓存稳定
- prefetch_models: 安装时预下载模型权重

app/models/task.py

@@ -14,6 +14,17 @@ def _now() -> datetime:
     return datetime.now(timezone.utc)
 
 
+# 任务状态枚举（单一真源，供 scheduler / cache_cleaner / views 引用）
+STATUS_UPLOADING = "uploading"   # 虚拟状态：分片上传中（不在 Task 表，由 UploadSession 映射）
+STATUS_QUEUED = "queued"
+STATUS_EXTRACTING = "extracting"
+STATUS_TRANSCRIBING = "transcribing"
+STATUS_SEGMENTING = "segmenting"
+STATUS_TRANSLATING = "translating"
+STATUS_DONE = "done"
+STATUS_FAILED = "failed"
+
+
 class Task(Base):
     __tablename__ = "task"
 
@@ -23,7 +34,7 @@ class Task(Base):
     # 视频在 upload_dir 下的相对路径（提取音频前后可能被删）
     source_path: Mapped[str] = mapped_column(String(1024), nullable=False)
     # 状态机：queued → extracting → transcribing → segmenting → translating → done | failed
-    status: Mapped[str] = mapped_column(String(32), nullable=False, default="queued")
+    status: Mapped[str] = mapped_column(String(32), nullable=False, default=STATUS_QUEUED, index=True)
     # 0-100 进度
     progress: Mapped[float] = mapped_column(Float, nullable=False, default=0.0)
     # 失败原因
@@ -33,7 +44,13 @@ class Task(Base):
     en_srt_path: Mapped[str | None] = mapped_column(String(1024), nullable=True)
     zh_srt_path: Mapped[str | None] = mapped_column(String(1024), nullable=True)
 
-    created_at: Mapped[datetime] = mapped_column(DateTime, default=_now)
+    # ---- 流水线阶段间传递的中间数据（scheduler 调度用）----
+    # ffmpeg 提取出的 wav 绝对路径（extract 阶段写，asr 阶段读）
+    wav_path: Mapped[str | None] = mapped_column(String(1024), nullable=True)
+    # ASR + 断句结果（asr 阶段写 JSON，translate 阶段读）。JSON 序列化的 list[Subtitle]
+    segments_json: Mapped[str | None] = mapped_column(Text, nullable=True)
+
+    created_at: Mapped[datetime] = mapped_column(DateTime, default=_now, index=True)
     updated_at: Mapped[datetime] = mapped_column(DateTime, default=_now, onupdate=_now)
 
     def __repr__(self) -> str:

app/models/upload_session.py

@@ -1,8 +1,9 @@
-"""分片上传会话 ORM：支撑断点续传。对齐 server 的 UploadSession 形态（SQLite 版）。"""
+"""分片上传会话 ORM：支撑断点续传。"""
 
 from __future__ import annotations
 
 import json
+import logging
 from datetime import datetime, timezone
 from typing import Any
 
@@ -11,6 +12,8 @@ from sqlalchemy.orm import Mapped, mapped_column
 
 from ..database import Base
 
+logger = logging.getLogger("audio2text.models")
+
 
 def _now() -> datetime:
     return datetime.now(timezone.utc)
@@ -26,7 +29,14 @@ class _IntList(TypeDecorator):
         return json.dumps(value) if value is not None else None
 
     def process_result_value(self, value: Any, dialect) -> list[int]:
-        return json.loads(value) if value else []
+        if not value:
+            return []
+        try:
+            return json.loads(value)
+        except (json.JSONDecodeError, TypeError):
+            # DB 脏数据（手工修改/并发写截断）不应让整行读取失败
+            logger.warning("uploaded_chunks JSON 解析失败，回退空列表：%r", value[:80])
+            return []
 
 
 class UploadSession(Base):
@@ -39,11 +49,13 @@ class UploadSession(Base):
     total_chunks: Mapped[int] = mapped_column(Integer, nullable=False)
     uploaded_chunks: Mapped[list[int]] = mapped_column(_IntList, default=list)
     # pending → completed（complete 成功）| abandoned（reaper 清理）
-    status: Mapped[str] = mapped_column(String(32), nullable=False, default="pending")
+    status: Mapped[str] = mapped_column(String(32), nullable=False, default="pending", index=True)
     # 拼接完成后的视频相对 upload_dir 路径
     final_path: Mapped[str | None] = mapped_column(String(1024), nullable=True)
-    # complete 后关联的 Task.id（直接引用，避免反向查找 source_path）
-    task_id: Mapped[int | None] = mapped_column(ForeignKey("task.id"), nullable=True)
+    # complete 后关联的 Task.id。ondelete=SET NULL：Task 被删时 session 保留（task_id 置空）
+    task_id: Mapped[int | None] = mapped_column(
+        ForeignKey("task.id", ondelete="SET NULL"), nullable=True
+    )
 
-    created_at: Mapped[datetime] = mapped_column(DateTime, default=_now)
+    created_at: Mapped[datetime] = mapped_column(DateTime, default=_now, index=True)
     updated_at: Mapped[datetime] = mapped_column(DateTime, default=_now, onupdate=_now)