fix(sse): initial ping + periodic heartbeat unblocks "connecting" state

SSE clients (agents_dashboard) consider the stream connected only after
receiving the first byte of body. The previous implementation flushed
headers and then blocked waiting for status diffs (sse_status) or log
lines (sse_agents_logs) — which could be silent for minutes. UI sat
on "connecting" indefinitely.

Fix:
- After WriteHeader + Flush, emit ":ping\n\n" comment (SSE spec, valid
  no-op) and flush. Unblocks client fgets immediately → state flips
  to "connected" in < 1s.
- Add 15s ticker emitting ":ping\n\n" so idle streams stay alive
  through Traefik / CDN proxies and clients detect dead servers.
- Same treatment for /sse/status and /sse/agents/{id}/logs (tail.go).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

internal/api/tail.go

@@ -47,15 +47,23 @@ func tailLogFile(ctx context.Context, path string, w http.ResponseWriter, flushe
 		}
 	}
 
-	// Tail the file: poll for new bytes every 200ms
+	// Tail the file: poll for new bytes every 200ms.
+	// Separate heartbeat ticker keeps proxies / clients alive on idle logs.
 	ticker := time.NewTicker(200 * time.Millisecond)
 	defer ticker.Stop()
+	heartbeat := time.NewTicker(15 * time.Second)
+	defer heartbeat.Stop()
 
 	reader := bufio.NewReader(f)
 	for {
 		select {
 		case <-ctx.Done():
 			return
+		case <-heartbeat.C:
+			if _, err := fmt.Fprint(w, ": ping\n\n"); err != nil {
+				return
+			}
+			flusher.Flush()
 		case <-ticker.C:
 			for {
 				line, err := reader.ReadString('\n')