replay: delay per complete frame instead of per chunk

ReplayPacer (MsgListener+MsgTalker) sits between packetizer output and
downstream listeners; sleeps after each complete NMEA/UBX frame before
forwarding. FileBackend no longer owns the delay — it reads as fast as
possible and the pacer provides back-pressure on the same thread.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

gui/data_model.py

@@ -153,6 +153,7 @@ class ReceiverManager(QObject):
         from listeners import NavSatQtListener, GsaQtListener
         from backend.serial_backend import SerialBackend
         from backend.file_backend   import FileBackend
+        from msg.replay_pacer       import ReplayPacer
 
         config = self._configs.get(rid)
         state  = self._states.get(rid)
@@ -168,7 +169,7 @@ class ReceiverManager(QObject):
         elif config.conn_type == 'serial':
             backend = SerialBackend(config.serial_port, config.baud_rate)
         else:
-            backend = FileBackend([config.file_path], delay=config.delay)
+            backend = FileBackend([config.file_path])
 
         backend.register_xcvr(xcvr)
 
@@ -177,10 +178,19 @@ class ReceiverManager(QObject):
         xcvr.register_listener(nmea_pkt)
         xcvr.register_listener(ubx_pkt)
 
-        ubx_pkt.register_listener(
+        if config.conn_type == 'file' and config.delay > 0:
+            nmea_sink = ReplayPacer(config.delay)
+            ubx_sink  = ReplayPacer(config.delay)
+            nmea_pkt.register_listener(nmea_sink)
+            ubx_pkt.register_listener(ubx_sink)
+        else:
+            nmea_sink = nmea_pkt
+            ubx_sink  = ubx_pkt
+
+        ubx_sink.register_listener(
             NavSatQtListener(lambda sats, r=rid: self._on_sat(r, sats))
         )
-        nmea_pkt.register_listener(
+        nmea_sink.register_listener(
             GsaQtListener(rid, lambda p, t, r=rid: self._on_pdop(r, p, t))
         )