fix(copaw-worker): restore model preflight + startup health writes (#1077 regression)

[LUOSENGWA]

Fixes the 18 baseline copaw/tests failures left behind by the #1077 Worker rewrite.

Problem A: test_worker_health.py — 15 failures

The tests patch copaw_worker.worker.check_model_service, but #1077's rewrite
of worker.py dropped the model-preflight import and the whole startup
health-state mechanism (e2fce4a had it; upstream did not preserve it).

Production gap fixed

Readiness previously hard-coded model: "validated at startup" — the model
component was never actually checked, so a Worker stayed "ready" while the
model provider was down (readiness false-positive). Now:

• start() runs a real model preflight (records health + Matrix notify on
  failure, without blocking boot)
• build_worker_readiness() re-checks model / copaw / matrix live on every poll
• start() records sync / matrix / bridge health in health.json
• Worker API liveness/readiness server starts during startup
  (build_worker_liveness / build_worker_readiness methods)

Problem B: test_sync.py — 3 failures

• _get_worker_info() requires the agt CLI, which is not available in a
  bare test container (real worker images copy it) — tests now monkeypatch it
• fake_mc did not write local files, so the post-mirror openclaw.json
  existence check failed — fake_mc now writes the file

Baseline comparison (full copaw/tests run)

	failed	passed
before (124f06d)	84	126
after	66	144

Delta == exactly the 18 fixed (15 worker_health + 3 sync), zero new
failures. All 46 tests in worker_health/sync/health now pass (0.34s).

Note

The remaining 66 failures (test_taskflow_tool / test_worker_matrix_channel /
test_worker_sync) are pre-existing upstream baseline failures unrelated to
this change; they are tracked separately.

问题 A：test_worker_health.py — 15 个失败

测试 patch 了 copaw_worker.worker.check_model_service，但 #1077 重写 worker.py 时丢掉了 model preflight 的 import 和整套启动期 health-state 机制（e2fce4a 中曾存在，上游未保留）——功能回归。

修复的生产缺口

修复前 readiness 硬编码 model: "validated at startup"——model 组件从不真正检查，model 服务挂了 Worker 仍显示 ready（readiness 假阳性）。现在：

• start() 执行真实 model preflight（写入 health + 失败时 Matrix 通知，但不阻塞启动）
• build_worker_readiness() 每次轮询时对 model / copaw / matrix 做实时检查
• start() 将 sync / matrix / bridge 组件健康状态写入 health.json
• Worker API 的 liveness/readiness 服务改为启动期创建（build_worker_liveness / build_worker_readiness 方法）

问题 B：test_sync.py — 3 个失败

• _get_worker_info() 依赖 agt CLI，裸测试容器里没有（真实 Worker 镜像有）——测试改为 monkeypatch
• fake_mc 不写本地文件，导致 mirror 后 openclaw.json 存在性检查失败——fake_mc 补上文件写入

基线对比（copaw/tests 全量）

	failed	passed
修复前（124f06d1）	84	126
修复后	66	144

差值 = 恰好修复 18 个（15 worker_health + 3 sync），零新增失败。worker_health/sync/health 三个文件共 46 个测试全部通过（0.34s）。

说明

剩余 66 个失败（test_taskflow_tool / test_worker_matrix_channel / test_worker_sync）是修复前就存在的上游基线失败，与本改动无关，另行跟踪。

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.