CI: mount tests leak background writers past TempDir cleanup

[khaliqgant]

Bug

The full Go contract suite can finish mount tests before their background watcher/writeback goroutines have stopped. Go then races t.TempDir() cleanup against new writes and fails with unlinkat ...: directory not empty.

Hosted evidence

Run: https://github.com/AgentWorkforce/relayfile/actions/runs/34015241241/job/101437753162

Two independent failures occurred in one run:

• cmd/relayfile-cli: TestMountLoopStartsWatcherBeforeBlockedInitialBootstrap
• internal/mountsync: TestHandleLocalChangesBatchesElevenFilesAndDefersPendingReceipts

Both failed only during testing.go TempDir removal with directory not empty; surrounding package tests otherwise ran to completion. This is unrelated to the workflow-only changes in relayfile#468 and is consistent with a background lifecycle leak.

Expected

Tests cancel and join every watcher, daemon, timer, and writeback goroutine before returning, so TempDir cleanup is deterministic.

Acceptance

• Add lifecycle regressions that fail if a writer survives test teardown.
• Ensure all affected goroutines/processes expose and use a join/termination boundary.
• Run the affected tests repeatedly and with -race where practical.
• go test ./... passes repeatedly in hosted Linux CI without rerunning.