Flaky test: TestDeleteHandlerTombstone fails on Windows

[hongliangl]

Describe the bug

TestDeleteHandlerTombstone in pkg/agent/controller/bgp fails on the Windows runner. The test does not report an assertion failure, the whole test binary is killed while its goroutines are parked in a testing/synctest bubble which never quiesces:

goroutine 2689 gp=0xc000718780 m=nil [synctest.Run (durable), synctest bubble 2]:
internal/synctest.Run(0xc00012a880)
	src/runtime/synctest.go:218
testing/synctest.Test(0xc000176800, 0xc000167b20)
	src/testing/synctest/synctest.go:290
antrea.io/antrea/v2/pkg/agent/controller/bgp.TestDeleteHandlerTombstone.func11(0xc000176800)
	pkg/agent/controller/bgp/controller_test.go:2726
...
FAIL	antrea.io/antrea/v2/pkg/agent/controller/bgp	4.769s

The other goroutines of the same bubble are parked on sync.WaitGroup.Wait (durable), chan receive (durable) and select (durable).

Occurrences

Same test, same failure mode, on unrelated PRs (#7803 only changes pkg/agent/route, pkg/agent/client and docs):

• 2026-08-13 https://github.com/antrea-io/antrea/actions/runs/31668396076/job/94347910290
• 2026-08-14 https://github.com/antrea-io/antrea/actions/runs/31758441710/job/94639410696
• 2026-08-18 https://github.com/antrea-io/antrea/actions/runs/32091404097/job/95574211338

Additional context

Only the Windows job is affected, the Linux one passes. The runner uses Go 1.26.5.

#8192 fixed a different flake in the same test file, so the remaining one may have the same root cause.

Note that testing/synctest does not treat locking a mutex as durably blocking, so a bubble containing a goroutine which is waiting for one never quiesces.