Fix intermittent test: TestNotifier/MultipleSubscribersStress (#865)

Try to fix one of the most commonly failing intermittent tests we have
left, which is the stress test in the notifier [1].

    --- FAIL: TestNotifier (0.00s)
        --- FAIL: TestNotifier/MultipleSubscribersStress (0.98s)
            notifier_test.go:409: Generated schema "notifier_2025_04_26t20_08_50_schema_04" with migrations [1 2 3 4 5 6] on line "main" in 376.579906ms [4 generated] [0 reused]
            notifier_test.go:434: Sending notification on "test_topic1": msg0
            notifier_test.go:434: Sending notification on "test_topic1": msg1
            notifier_test.go:434: Sending notification on "test_topic1": msg2
            notifier_test.go:434: Sending notification on "test_topic1": msg3
            notifier_test.go:434: Sending notification on "test_topic1": msg4
            notifier_test.go:434: Sending notification on "test_topic1": msg5
            notifier_test.go:434: Sending notification on "test_topic1": msg6
            notifier_test.go:472: Channel  0 contains   5 message(s)
            notifier_test.go:472: Channel  1 contains   7 message(s)
            notifier_test.go:472: Channel  2 contains   0 message(s)
            notifier_test.go:477:
                    Error Trace:	/home/runner/work/river/river/internal/notifier/notifier_test.go:477
                    Error:      	Should NOT be empty, but was 0xc000244770
                    Test:       	TestNotifier/MultipleSubscribersStress
            riverdbtest.go:277: Checked in schema "notifier_2025_04_26t20_08_50_schema_04"; 1 idle schema(s) [4 generated] [2 reused]
    FAIL
    FAIL	github.com/riverqueue/river/internal/notifier	2.025s

Rebuild the test a little so that each listen goroutine waits for at
least one message to come through instead of only an arbitrary sleep.
The sleep ~always passes locally, but it seems to be that in CI it's
possible for the sending goroutine to be entirely parked around when a
listen/unlisten loop is happening.

[1] https://github.com/riverqueue/river/actions/runs/14684674639/job/41211780010

Author: brandur

internal/notifier/notifier_test.go

@@ -459,6 +459,13 @@ func TestNotifier(t *testing.T) {
 					// Pause a random brief amount of time.
 					serviceutil.CancellableSleep(ctx, randutil.DurationBetween(15*time.Millisecond, 50*time.Millisecond))
 
+					// Wait for at least one message to come through. This
+					// generally isn't necessary, but we can run into strange
+					// problems in slow environments like GitHub Actions where
+					// the send goroutine can apparently be paused entirely
+					// around the sleep above between listen and unlisten.
+					riversharedtest.WaitOrTimeout(t, notifyChan)
+
 					sub.Unlisten(ctx)
 				}
 			}()
@@ -469,12 +476,12 @@ func TestNotifier(t *testing.T) {
 		<-sendNotificationsDone          // wait for notifications goroutine to finish
 
 		for i := range notifyChans {
-			t.Logf("Channel %2d contains %3d message(s)", i, len(notifyChans[i]))
-
 			// Don't require a specific number of messages to have been received
 			// since it's non-deterministic, but every channel should've gotten
-			// at least one message. It my test runs, they receive ~15 each.
-			require.NotEmpty(t, notifyChans[i])
+			// at least one message (as measured by the WaitOrTimeout above).
+			// In my test runs, they receive ~15 each. The WaitOrTimeout
+			// consumes one message, so we add +1 when measuring each channel.
+			t.Logf("Channel %2d contains %3d message(s)", i, len(notifyChans[i])+1)
 		}
 	})