fix(engine): giam jitter bridge timer va fade-out khi transport stall

- RealtimeFxLoop: timeBeginPeriod(1) + sleep_until thay Sleep(1) -> loopMax 32.9->24.8ms
- plugins.py _pump_output: stall-fade 8 block gain 1->0 khi gap > 5.5*block_sec (~29ms),
  burst send, fade-in 0->1 khi resume; poll dung asyncio.to_thread(time.sleep) vi asyncio.sleep
  bi cap 15.6ms tren Win11 (IOCP) du timeBeginPeriod
- fx_realtime.py: stats dem publish/drop/stale/read cho diagnostics
- rebuild bridge binary

app/core/fx_realtime.py

@@ -128,6 +128,9 @@ class FxRealtimeSession:
         # Phase 2.8: C++ consume qua SHM control ring.
         self.pending_params = {}   # {slot_idx: {key: value}}
         self.latencies = {}        # {slot_idx: samples}
+        # Diagnostics (FXRT crackle hunt): drop/publish counters.
+        self.stats = {'publish': 0, 'drop_batch': 0, 'drop_blocks': 0,
+                      'stale_flush': 0, 'read_blocks': 0, 'read_empty': 0}
         self._spawn()
 
     # ── process ─────────────────────────────────────────────────────────────
@@ -266,13 +269,16 @@ class FxRealtimeSession:
         self._in_pending = []
         self._in_pending_t0 = None
         avail = h.in_slots - (h.in_write - h.in_read)
+        self.stats['publish'] += len(frames)
         if avail <= 0:
             h.in_read += len(frames)  # ring full → drop cả batch
+            self.stats['drop_batch'] += len(frames)
             return
         n = min(len(frames), avail)
         drop = len(frames) - n
         if drop > 0:
             h.in_read += drop  # drop frame cũ nhất
+            self.stats['drop_blocks'] += drop
         for i in range(n):
             L, R = frames[i + drop]
             slot = (h.in_write + i) & (h.in_slots - 1)
@@ -287,6 +293,7 @@ class FxRealtimeSession:
         burst) → không kẹt latency. Gọi từ _pump_output mỗi vòng."""
         with self._lock:
             if self._in_pending and self._in_pending_t0 is not None and time.time() - self._in_pending_t0 > timeout:
+                self.stats['stale_flush'] += 1
                 self._flush_input_locked()
 
     def read_output(self):
@@ -303,6 +310,9 @@ class FxRealtimeSession:
                 inter[1::2] = self._np[br:br + self.block_size]
                 out.append(inter.tobytes())
                 h.out_read += 1
+        self.stats['read_blocks'] += len(out)
+        if not out:
+            self.stats['read_empty'] += 1
         return out
 
     def heartbeat_age(self):