feat(observability): 接入 Prometheus 指标与 OpenTelemetry/Tempo 链路

[gac0812]

Summary

• 后端接入 Prometheus 指标（QPS、延迟、连接数、工具调用、外部依赖、数据库）和 OpenTelemetry → Tempo 链路（每轮对话、工具、ASR/LLM/TTS 耗时）。
• Grafana 只是看数据的界面，不是这次交付目标；代码改 HTTP / WebSocket / Agent / 外部 SDK，是为了打点和打 span，不改业务功能。
• GET /metrics 供 Prometheus scrape；配置 TIMEFLOW_OTEL_EXPORTER_OTLP_ENDPOINT 后才导出 trace，未配置则 tracing 关闭。
• 指标 label 与 span attribute 使用封闭枚举，不写 transcript、session id、坐标或凭证。

Closes #358

Test plan

• GET /metrics 返回 Prometheus text，且不把 /metrics 自己算进业务 HTTP QPS
• HTTP / WebSocket / 语音阶段 / 工具 / 外部依赖 / 数据库连接池都有对应指标
• 一轮对话结束后，Tempo 能立刻看到 voice.turn 以及嵌套的 tool / ASR / LLM / TTS / DB span，不必等 WebSocket 断开
• 本地 Grafana 能画出折线图和 Trace 表（效果图见 feat(observability): 接入 Prometheus 指标与 OpenTelemetry/Tempo 链路 #358）
• 业务行为不变：鉴权、快照、提醒、语音对话语义未改

Verification

cd backend
bash scripts/check.sh

本地结果：1090 passed, 49 skipped，coverage 96.73%（门槛 95%）。

覆盖要点：

• HTTP 中间件按路由模板计数；/metrics 不计入业务 QPS
• voice.turn 单独成 trace，工具 span 保持调用顺序
• 未知工具名折叠为 other；失败 turn 标记 ERROR
• settings 在本地 .env 有 Tempo 地址时仍能断言 OTLP 开关

效果对照

Grafana 把 Prometheus / Tempo 画出来即可，截图不进源码树：

Overview + HTTP

WebSocket + Voice / tools

External / database

Traces（Tempo）

单轮 voice.turn 瀑布图

[gac0812]

已补上服务端会话数据埋点（占用阶段 Gauge + 挂断原因 ui_hangup / idle_timeout / tool_end）。session_id 只留进程内，不进 Prometheus label。

打断行为拆到后续草稿 PR，不放在本 PR。

[codecov]

Codecov Report

❌ Patch coverage is 95.87799% with 50 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...w/infrastructure/external/llm/openai_compatible.py	82.14%	9 Missing and 1 partial ⚠️
...ackend/src/timeflow/intelligence/composed/agent.py	80.00%	5 Missing and 3 partials ⚠️
...flow/infrastructure/external/tts/qwen_audio_tts.py	75.00%	6 Missing ⚠️
.../timeflow/infrastructure/observability/database.py	93.10%	2 Missing and 4 partials ⚠️
...w/infrastructure/external/location/tencent_maps.py	86.95%	2 Missing and 1 partial ⚠️
...low/infrastructure/external/realtime/qwen_audio.py	91.42%	2 Missing and 1 partial ⚠️
...ackend/src/timeflow/intelligence/realtime/agent.py	86.36%	3 Missing ⚠️
...c/timeflow/intelligence/realtime/schedule_tools.py	72.72%	3 Missing ⚠️
backend/src/timeflow/gateway/websocket/endpoint.py	96.61%	1 Missing and 1 partial ⚠️
...eflow/infrastructure/external/asr/qwen_realtime.py	91.66%	2 Missing ⚠️
... and 3 more

Flag	Coverage Δ
backend	96.57% <95.87%> (-0.17%)	⬇️
frontend	89.80% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
backend/src/timeflow/composition.py	97.14% <100.00%> (+0.08%)	⬆️
backend/src/timeflow/data/database.py	100.00% <100.00%> (ø)
backend/src/timeflow/data/repositories/account.py	95.34% <100.00%> (+0.34%)	⬆️
backend/src/timeflow/gateway/http/auth.py	95.86% <100.00%> (+0.29%)	⬆️
...ackend/src/timeflow/gateway/http/reminder_state.py	100.00% <100.00%> (ø)
...end/src/timeflow/gateway/http/schedule_snapshot.py	100.00% <100.00%> (ø)
...end/src/timeflow/gateway/observability/__init__.py	100.00% <100.00%> (ø)
backend/src/timeflow/gateway/observability/http.py	100.00% <100.00%> (ø)
...end/src/timeflow/gateway/observability/sessions.py	100.00% <100.00%> (ø)
...nd/src/timeflow/gateway/observability/websocket.py	100.00% <100.00%> (ø)
... and 25 more

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[fennoai]

Reviewed the complete observability diff, including HTTP/WebSocket lifecycle, agent/provider instrumentation, database hooks, settings, and the new tests. The implementation is generally cohesive, but the username enrichment path can make otherwise successful voice turns fail when the database is temporarily unavailable. python3 -m compileall -q backend/src backend/tests and git diff --check pass; backend/scripts/check.sh could not run because uv is not installed in the review environment.