[v0.1.x-branch] Backport #966: waved: fix mailbox-compat test tempdir cleanup flake under -race

[github-actions]

Backport of #966

---

Problem

The unit-race job intermittently fails with:

--- FAIL: TestStartMailboxIngressConcurrentIncompatible
    testing.go:1464: TempDir RemoveAll cleanup: unlinkat .../001: directory not empty

(seen e.g. in the run for #895, but the test and code are on main and the
flake is not reorg-related). Despite running under -race, this is not a
data-race report — Go's t.TempDir() calls t.Errorf when its RemoveAll
fails, which fails the test.

Root cause

TestStartMailboxIngressConcurrentIncompatible and
TestStartMailboxIngressAlreadyIncompatible start the durable serverconn
egress actor with StartEgress() but tore it down with a deferred,
fire-and-forget runtime.Stop(). DurableActor.Stop() only cancels the
actor context and returns immediately (a.stopOnce.Do(func(){ a.cancel() }));
only StopAndWait/Wait block on a.done.

So the teardown ordering was:

1. defer Stop() — signals cancel, returns instantly
2. t.Cleanup: DB.Close()
3. t.Cleanup: os.RemoveAll(tempdir)

The egress goroutine kept issuing queries against the shared SQLite handle
during steps 2–3. An in-flight connection re-materializes the WAL/-shm
sidecar files, so RemoveAll's final rmdir races and fails with
directory not empty.

Fix

Switch both tests to StopAndWait(context.Background()) in the deferred
teardown so the egress goroutine has fully drained before the DB is closed
and the temp dir removed. Test-only change.

Verification

• Before: -count=300 -cpu=4 -race failed dozens of times.
• After: passes cleanly across 1000+ iterations.
• go vet, make fmt-changed, make lint-changed-local (0 issues) all pass.

Note (separate, pre-existing)

While stressing this I observed a distinct, much rarer data race originating
at baselib/actor/durable_actor.go:508 (the egress worker goroutine) — ~1 in
several thousand iterations, identical on main and the reorg branches and
untouched by this change. It is orthogonal to the reported cleanup flake and
worth a separate investigation; I could not reliably reproduce it to capture a
full trace.