fix: prevent auto-next double-fire and unblock end-of-queue stop

Two bugs found in auto-next state management:

Bug A - Double-fire: check_auto_next() could fire twice for the same
track end because state.update(TRANSITIONING) is async and hasn't
propagated by the next polling cycle. The _auto_next_in_flight flag
was already set synchronously but check_auto_next never checked it.
Fix: add _auto_next_in_flight guard at the top of check_auto_next().

Bug B - End-of-queue stop blocked: when auto-next fires for the last
track in queue, next() detects end-of-queue and calls stop(), but
stop() was blocked by the _auto_next_in_flight guard (which can't
distinguish stale Plex stops from legitimate internal stops).
Fix: clear _auto_next_in_flight before calling stop() in next()'s
end-of-queue branches (offset<0 and offset>=total_count).

Author: aquantumofdonuts

plex/adapters.py

@@ -689,6 +689,8 @@ def check_auto_next(self, changed: DotMap):
             return False
         if self._suppress_auto_next:
             return False
+        if self._auto_next_in_flight:
+            return False
         if self._active_operation_id:
             return False
         if self.queue is None:
@@ -1072,10 +1074,12 @@ async def next(self, revert=False):
 
         if current_offset < 0:
             logger.debug("%s next guard stop: offset<0 current=%s start=%s", self.dlna.name, current_offset, start_offset)
+            self._auto_next_in_flight = False
             await self.stop()
             return
         if not math.isinf(total_count) and current_offset >= total_count:
             logger.debug("%s next guard stop: offset>=total current=%s total=%s", self.dlna.name, current_offset, total_count)
+            self._auto_next_in_flight = False
             await self.stop()
             return
         self.state.update(state="TRANSITIONING")