fix: stabilize OpenClaw CLI insight context

[akaDRJ]

Summary

This fixes a false-negative connection-health path in the control center when invoking openclaw CLI commands from the dashboard runtime.

What changed

• run OpenClaw CLI calls from the OpenClaw workspace root instead of the repo cwd
• strip npm_* / INIT_CWD pollution from child-process env before invoking openclaw
• avoid injecting OPENCLAW_* overrides that can break plugin/config resolution on real installs
• raise CLI insight timeouts to better match slower local runtimes
• make expired insight caches await a fresh value instead of immediately serving stale blocked state
• add regression tests for workspace-root resolution and CLI env sanitization

Why

On a real OpenClaw installation, the Settings → Connection health card could incorrectly show:

• Gateway is not reachable
• openclaw.json is missing or invalid
• Runtime status is still loading

…even though the local OpenClaw runtime was healthy.

The root causes were a mix of:

• invoking openclaw from the wrong cwd
• inheriting npm runner env pollution
• stale cache behavior returning outdated blocked summaries on the first refresh after expiry
• timeout budgets that were too aggressive for slower local CLI responses

Validation

• npm test → 106/106 pass
• verified Settings → Connection health renders healthy Gateway / Config / Runtime on a real local install
• verified LAN-access local config separately (UI_BIND_ADDRESS=0.0.0.0) without including that local-only change in this PR

[TianyiDataScience]

谢谢你先把这条 draft 放出来，方向我认同。

我这边先记下来了，不过因为它现在还是 draft，我就先不急着动 merge。等你基于最新 main rebase 完、切成 ready for review 之后，我再按最新主干一起看。