fix(test): stabilize watcher healthy-peer restart fixture

[Amplify-Logic]

Summary

• Fix the flaky tests/fm-watcher-lock.test.sh healthy-peer restart case that timed out at 80s (status 124) on unmodified main.
• Root cause was a test-fixture race, not a product hang in the arm/restart path: the Node peer was launched with a SIGTERM no-op handler, but the test proceeded as soon as the shell recorded the peer pid. On macOS, restart can deliver SIGTERM before Node finishes installing that handler; default TERM then kills the peer, so restart falls into the child-owned attach/replace path and blocks until the suite's 80s wait deadline.
• Evidence: the failing assertion was restart did not exit zero after reporting healthy peer (status 124) — 124 is the fixture timeout, not an assertion mismatch. Reproduced deterministically when the peer was still in startup (handler not installed). After waiting for an explicit readiness marker written only after the SIGTERM handler is installed, restart correctly reports the healthy peer and exits 0 without attaching.
• Fixture now waits for that readiness file (failing fast if the handler never installs) and cleans up peer/arm processes on assertion failure so leftovers cannot poison later cases. Product restart behavior is unchanged.

Test plan

• bash tests/fm-watcher-lock.test.sh — at least 5 consecutive green runs in isolation on the task machine (predecessor + 2 reconfirm runs after recovery)
• Confirm CI green on this PR
• No other suites expected to change; this is fixture-only