feat(ai): AI 教练 —— 晨间简报、运动处方、趋势归因与 Copilot

数值全部在服务端算好再交给模型，模型只做解读。让模型从 CSV 里自己推
z 分数，它算错的次数足以让简报引用图表反驳它的数字。

- services/insights.py：z 分数（28 天个人基线，且**排除当天**——用一个
  值参与算出来的均值去衡量它自己，会把真实离群点摊平）、13 个月趋势斜率
  （按序数日期最小二乘，手表放充电器上一周不会压缩 x 轴）、近 7 天活动量
  对比。
- services/coach.py：三套提示词 + 回复解析，每套都配一个规则引擎版本。
  网关一次生成要几分钟，上游被限流时给一个朴素的答案，好过给一张空卡片。
- services/ai.py：多轮 chat()、SSE stream()、complete()/stream_chat()，
  以及 extract_json()——上游是推理模型，可见输出以思维链开头，所以从末尾
  倒着找最后一个配平的 JSON（字符串感知，扛得住引号里的 } 和转义引号）。
- 接口 briefing / trend-insight / copilot(SSE)，缓存表 ai_insights。
- 前端：今日页晨报卡（后台生成 + 轮询升级）、全局 Copilot 浮窗、指标详情
  页归因面板。features.ai 打开。

实测（对着自建 ai-gateway）：晨报一次 273 秒，缓存命中 18 毫秒——所以简报
绝不能同步阻塞首屏。网关的流式通道比阻塞通道更不可靠：同一条提示词流式
139 秒后返回「所有模型均不可用」，阻塞则成功，因此 stream_chat() 在流式零
输出时对同一模型退回非流式重试。Copilot 实测 TTFB 9ms、全程 40 秒。

顺带修两处：refresh 原来只跳过缓存读、不删行，导致「重新生成」后的轮询读
到旧行、看到 cached 就停了，用户一直盯着他刚要求替换掉的那段字；基线零方差
时原来返回 z=0.0，把「和每一条观测都不同」标成「完全正常」，改为 z=null。

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/.env.example

@@ -53,7 +53,9 @@ CORS_ORIGIN=http://localhost:3000,http://localhost:5173
 # absorbs single-vendor quota limits. Reached directly, bypassing any local
 # HTTP proxy. NOTE: its NVIDIA upstream is a large reasoning model — replies
 # can take 2-3 minutes, so set AI_TIMEOUT_SECONDS accordingly.
-AI_GATEWAY_BASE_URL=http://129.146.26.249:5100/v1
+# HTTPS (Caddy, strips the /ai prefix) rather than http://…:5100 — the token
+# rides in an Authorization header and should not cross the internet in clear.
+AI_GATEWAY_BASE_URL=https://oracle.zichuan.xyz/ai/v1
 AI_GATEWAY_TOKEN=
 AI_GATEWAY_MODEL=ai-gateway-auto
 
@@ -73,7 +75,17 @@ AI_MODEL_CHAIN=gateway,gemini-flash,llama-70b
 # payload always fits that model's own context window.
 AI_DAY_BUDGET=365
 
-AI_TIMEOUT_SECONDS=180
+# Measured against the gateway, not guessed: a trivial prompt took 138s end to
+# end, because its primary upstream emits a full chain of thought before the
+# answer. Nothing user-facing blocks on this (the briefing generates in a
+# background thread), but the timeout still has to clear the real latency.
+AI_TIMEOUT_SECONDS=300
 # Output cap. Reasoning models spend part of it thinking before they answer;
 # entries that need more declare their own budget in services/ai.py.
 AI_MAX_TOKENS=1024
+
+# Output cap for the AI coach (晨报 / 趋势归因 / Copilot). Larger than
+# AI_MAX_TOKENS above: the same reasoning trace is spent from this budget
+# before the answer starts, and at 1024 the reply was all thinking with the
+# JSON truncated away.
+AI_COACH_MAX_TOKENS=4000