Refactor FSEvents handling in serve_util.rs for improved reliability

- Removed unnecessary sleep duration for FSEvents on macOS, reducing it from 500ms to 100ms to enhance responsiveness while still allowing time for event processing.
- Updated comments to clarify the rationale behind the timing adjustments and noted the need for a more robust solution in the future.

tests/rojo_test/serve_util.rs

@@ -38,14 +38,6 @@ pub fn run_serve_test(test_name: &str, callback: impl FnOnce(TestServeSession, R
     let mut session = TestServeSession::new(test_name);
     let info = session.wait_to_come_online();
 
-    // On macOS, the FSEvents-backed file watcher (via notify-debouncer-full)
-    // may not have fully populated its internal cache by the time the HTTP
-    // server reports as online. Without a warm cache, the debouncer cannot
-    // correctly classify events (e.g., it may miss a Remove if the file was
-    // not yet in the cache). Give it time to complete its initial scan.
-    #[cfg(target_os = "macos")]
-    thread::sleep(Duration::from_secs(1));
-
     redactions.intern(info.session_id);
     redactions.intern(info.root_instance_id);
 
@@ -99,9 +91,11 @@ impl TestServeSession {
         // This is an ugly workaround for FSEvents sometimes reporting events
         // for the above copy operations, similar to this Stack Overflow question:
         // https://stackoverflow.com/questions/47679298/howto-avoid-receiving-old-events-in-fseventstream-callback-fsevents-framework-o
-        // 500ms gives FSEvents time to flush any stale events from the copy.
+        // We'll hope that 100ms is enough for FSEvents to get whatever it is
+        // out of its system.
+        // TODO: find a better way to avoid processing these spurious events.
         #[cfg(target_os = "macos")]
-        std::thread::sleep(Duration::from_millis(500));
+        std::thread::sleep(Duration::from_millis(100));
 
         let port = get_port_number();
         let port_string = port.to_string();