fix(stream): goroutine handoff to prevent post-SyncResponse notification drops

[steiler]

Fixes #440.

Summary

• buildTreeSyncWithDatastore previously called syncToRunning inline, blocking the consumer goroutine for the entire duration of ApplyToRunning → performRevert. Notifications arriving during this window waited up to notificationSendTimeout (5 s) and were then silently and permanently dropped, leaving the Running view of the device incomplete until the datastore was recreated.
• The CI failure in StreamSync: post-SyncResponse notifications silently dropped when syncToRunning blocks the consumer goroutine #440 (mandatory password leaf absent for 34+ minutes) was caused exactly by this: a post-SyncResponse onChange notification was dropped while performRevert held the goroutine.

Changes

pkg/datastore/target/gnmi/stream.go

• On SyncResponse and ticker events, snapshot the local syncTree, reset it to a fresh empty tree, and hand syncToRunning off to a background goroutine. The main select loop never blocks on a commit and keeps draining updChan at all times.
• atomic.Bool syncRunning — ticker skips if a goroutine is already in flight (skip-if-busy, not queue).
• tickerChan starts as nil (never fires in Go select) and is activated lazily on the first SyncResponse, replacing the previous tickerActive bool flag — removes one redundant variable.
• syncToRunning signature simplified: no mutex parameter (each goroutine owns its tree exclusively after handoff), no tree return value.
• updChanBufSize and notifSendTimeout are configurable fields (defaults unchanged) to enable deterministic regression tests.
• NewEmptyTree failure after handoff exits the loop rather than continuing with a shared tree that the goroutine is still reading.

pkg/datastore/target/gnmi/stream_test.go (new)

• TestBuildTreeSyncWithDatastore_PostSyncNotificationsNotDropped — regression test for this bug. Uses a zero-buffer updChan and a 20 ms send timeout: with the old code the consumer goroutine is blocked so pool workers time out and the second ApplyToRunning is never called (test fails); with the fix the goroutine handoff keeps the loop draining and the ticker commits the post-sync notifications (test passes).
• TestBuildTreeSyncWithDatastore_NewEmptyTreeFailureExits — verifies the loop exits cleanly on NewEmptyTree failure.

docs/adr/0003-streamsync-goroutine-handoff-for-notification-drop-prevention.md (new)

• Records the decision, the alternatives considered (Options 2–5 from StreamSync: post-SyncResponse notifications silently dropped when syncToRunning blocks the consumer goroutine #440), and why Option 4 (per-window drain) was deferred.

[codecov]

Codecov Report

❌ Patch coverage is 80.85106% with 9 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pkg/datastore/target/gnmi/stream.go	80.85%	6 Missing and 3 partials ⚠️

📢 Thoughts on this report? Let us know!

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.