fix(stream): report stream failures via StreamError + retry transient stat

Stream-request failures previously reported status:"failed", which the web
treated as a download failure — it left the task unstreamable and surfaced a
misleading 20s timeout. Report them through a dedicated StreamError field
instead, so the web clears the stream flag and shows the real reason without
touching the download status.

- StatusUpdate gains StreamError (json: streamError)
- OnStreamRequested reports failures via a reportStreamError helper (path
  rejected, file not found, no video in dir) instead of status:"failed"
- os.Stat is retried 3× (300ms) before giving up — NFS can transiently fail
  (ESTALE/EAGAIN/timeout), the root of the intermittent "works on the 3rd try"
- dispatch OnStreamRequested off the sync loop (goroutine): it does blocking
  I/O (stat retries, ffprobe in SetFile) that would otherwise stall task
  dispatch + status reporting for other items

internal/agent/daemon.go

@@ -246,8 +246,12 @@ func (d *Daemon) Run(ctx context.Context) error {
 		}
 	}
 	d.sync.OnStreamRequest = func(req StreamRequest) {
+		// Off the sync loop: the handler does blocking I/O (os.Stat retries on
+		// NFS, then ffprobe in SetFile) — running it inline would stall task
+		// dispatch + status reporting for other items. The single-stream model
+		// (atomic SetFile swap, last-wins) tolerates concurrent requests.
 		if d.OnStreamRequested != nil {
-			d.OnStreamRequested(req)
+			go d.OnStreamRequested(req)
 		}
 	}
 	d.sync.OnStreamSession = func(sess StreamSession) {