lampOS: the second route's route checks give tool_runner #1984's policed argument, so lamp-voice's tests compile again on #1712

[deehw]

lamp-voice's tests compile again on #1712. One commit, abb07bdba, on #1712's 00f05a016.

What broke:

• lampOS: Jarvis's voice features checked on the second route: reports_stop and dance (held reports_stop fixed in #1925, reports_state on lamp-voice) #1944's second_provider/worker/route_checks.rs (merged at cb2d663b6) calls tool_runner(tools, None, None) four times, at lines 79, 85, 117 and 143.
• lampOS: the second route holds the lamp's actions for the Policy's decision about their turn, as Live does, with Live's journal lines, and ends a held turn in silence (#1712 item 6) #1984 (f837edd5a, merged just before) gave tool_runner a fourth argument, policed: bool.
• So since 05:07 every voice PR's ARM64 run onto lampOS: the second voice route asks DeepSeek, OpenRouter or Gemini directly, searches with DeepSeek or Exa, speaks with Gemini, records each request's times and is timed by probe-cascade, each off unless set (#1686's P3s) #1712 fails at compile, with E0061 and 4 errors (lampOS: probe-cascade --wire --kept times only 200s, counts the rest closed, tests its arms and takes at most 20 runs (#2044's review) #2051's 39333751988).

The fix: , false on those four calls, as the worker's own tests pass it (worker.rs:4131). These checks then run without the Policy's gate, as they did when #1944 was written. Nothing else changes.

Checked on this Mac:

• rustfmt: clean;
• Clippy, -p lamp-voice --all-targets -D warnings: clean, so the tests compile;
• lamp-voice tests, one thread: 2585 passed, 0 failed (13 ignored);
• workspace build: clean.

ARM64 runs on this pull request.

[deehw]

Reviewer, 05:35, head abb07bdba: right, no finding. Merge it first: every voice PR onto #1712 fails to compile until it's in. tool_runner (worker.rs:1808) now takes policed: bool from #1984, and the four calls in #1944's route_checks.rs get false, as the worker's own tests pass it, so the checks run without the Policy's gate, as written. My part in it: I checked #1944 merged onto #1712's head at 05:00, before #1984 landed, so my 05:04 'both tests pass there' was true of a head that #1984 then changed. Two PRs I read separately touched the same signature.