[flow] Suppress spurious EdenFS watcher error on clean shutdown

Summary:
Changelog: [internal]

During `flow stop`, the exit hook destroys the EdenFS watcher instance
before the listen loop is canceled. When Lwt yields during the 1-second
shutdown window, the listen loop wakes up and tries to call
get_changes_async on the destroyed instance, producing a spurious
"EdenfsWatcherInstance was previously destroyed" error that gets logged
to Scuba as EDENFS_WATCHER_ERROR.

Detect this specific error in the listen loop and treat it as a clean
shutdown (Lwt.Canceled) instead of an error. This prevents false error
events from polluting Scuba on every normal `flow stop`.

Reviewed By: SamChou19815

Differential Revision: D94991663

fbshipit-source-id: bbe0eef3165bea16de8a0eb2f59c1062062333b1

Author: panagosg7

src/hack_forked/stubs/edenfs_watcher/edenfs_watcher.ml

@@ -29,3 +29,5 @@ let is_available () = false
 let watch_spec _options = failwith "not implemented"
 
 let hooks_upon_clean_exit : (unit -> unit) list ref = ref []
+
+let is_instance_destroyed () = false

src/monitor/fileWatcher.ml

@@ -686,6 +686,11 @@ end = struct
             { MonitorProt.changed_mergebase = None; missed_changes = true };
         broadcast env;
         Lwt.return env
+      | Error _ when Edenfs_watcher.is_instance_destroyed () ->
+        (* The instance was destroyed by the exit hook during shutdown.
+           This is expected and not an error — treat it as a clean stop. *)
+        Logger.info "EdenFS watcher instance was destroyed (shutdown in progress)";
+        raise Lwt.Canceled
       | Error err ->
         let msg = Edenfs_watcher.show_edenfs_watcher_error err in
         let backtrace = Printexc.get_backtrace () in