test(slack): wait for the old link to drop before the restart reply

The restart spec closed the first runtime and started a second one at
once. The fake server lists a link until its handler sees the socket
close, so Connected was already true for the old link, and the reply
the spec pushed next went to the closing socket and was lost. The spec
failed on every run with one CPU and on about one run in three with
more.

Wait until the fake reports no link before the second runtime starts.
30 runs each at 1, 4 and 20 CPUs under -race now pass, also with six
copies running at once.

Assisted-by: Claude:claude-sonnet-5-5 go-test

addons/slack/transport/follow_test.go

@@ -131,6 +131,10 @@ var _ = Describe("which threads the bot follows", func() {
 		info, _ := os.Stat(path)
 		Expect(info.Mode().Perm()).To(Equal(os.FileMode(0o600)))
 		b.rt.Close() // the add-on stops
+		// The fake lists the old link until its handler sees the close. A
+		// reply pushed before that goes to the closing socket and is lost, and
+		// start would also take the old link for the new one.
+		Eventually(func() bool { return f.Connected(appToken) }, 5*time.Second).Should(BeFalse())
 
 		b2 := start(f, dir, "own")
 		f.Push(appToken, slacktest.Reply(channel, "U1", "after the restart", mention["ts"].(string)))