fix: guarantee END sentinel delivery when stream bridge queue is full (#1695)

When MemoryStreamBridge queue reaches capacity, publish_end() previously
used the same 30s timeout + drop strategy as regular events. If the END
sentinel was dropped, subscribe() would loop forever waiting for it,
causing the SSE connection to hang indefinitely and leaking _queues and
_counters resources for that run_id.

Changes:
- publish_end() now evicts oldest regular events when queue is full to
  guarantee END sentinel delivery — the sentinel is the only signal that
  allows subscribers to terminate
- Added per-run drop counters (_dropped_counts) with dropped_count() and
  dropped_total properties for observability
- cleanup() and close() now clear drop counters
- publish() logs total dropped count per run for easier debugging

Tests:
- test_end_sentinel_delivered_when_queue_full: verifies END arrives even
  with a completely full queue
- test_end_sentinel_evicts_oldest_events: verifies eviction behavior
- test_end_sentinel_no_eviction_when_space_available: no side effects
  when queue has room
- test_concurrent_tasks_end_sentinel: 4 concurrent producer/consumer
  pairs all terminate properly
- test_dropped_count_tracking, test_dropped_total,
  test_cleanup_clears_dropped_counts, test_close_clears_dropped_counts:
  drop counter coverage

Closes #1689

Co-authored-by: voidborne-d <voidborne-d@users.noreply.github.com>

backend/packages/harness/deerflow/runtime/stream_bridge/memory.py

@@ -25,6 +25,7 @@ class MemoryStreamBridge(StreamBridge):
         self._maxsize = queue_maxsize
         self._queues: dict[str, asyncio.Queue[StreamEvent]] = {}
         self._counters: dict[str, int] = {}
+        self._dropped_counts: dict[str, int] = {}
 
     # -- helpers ---------------------------------------------------------------
 
@@ -32,6 +33,7 @@ class MemoryStreamBridge(StreamBridge):
         if run_id not in self._queues:
             self._queues[run_id] = asyncio.Queue(maxsize=self._maxsize)
             self._counters[run_id] = 0
+            self._dropped_counts[run_id] = 0
         return self._queues[run_id]
 
     def _next_id(self, run_id: str) -> str:
@@ -48,14 +50,41 @@ class MemoryStreamBridge(StreamBridge):
         try:
             await asyncio.wait_for(queue.put(entry), timeout=_PUBLISH_TIMEOUT)
         except TimeoutError:
-            logger.warning("Stream bridge queue full for run %s — dropping event %s", run_id, event)
+            self._dropped_counts[run_id] = self._dropped_counts.get(run_id, 0) + 1
+            logger.warning(
+                "Stream bridge queue full for run %s — dropping event %s (total dropped: %d)",
+                run_id,
+                event,
+                self._dropped_counts[run_id],
+            )
 
     async def publish_end(self, run_id: str) -> None:
         queue = self._get_or_create_queue(run_id)
-        try:
-            await asyncio.wait_for(queue.put(END_SENTINEL), timeout=_PUBLISH_TIMEOUT)
-        except TimeoutError:
-            logger.warning("Stream bridge queue full for run %s — dropping END sentinel", run_id)
+
+        # END sentinel is critical — it is the only signal that allows
+        # subscribers to terminate.  If the queue is full we evict the
+        # oldest *regular* events to make room rather than dropping END,
+        # which would cause the SSE connection to hang forever and leak
+        # the queue/counter resources for this run_id.
+        if queue.full():
+            evicted = 0
+            while queue.full():
+                try:
+                    queue.get_nowait()
+                    evicted += 1
+                except asyncio.QueueEmpty:
+                    break  # pragma: no cover – defensive
+            if evicted:
+                logger.warning(
+                    "Stream bridge queue full for run %s — evicted %d event(s) to guarantee END sentinel delivery",
+                    run_id,
+                    evicted,
+                )
+
+        # After eviction the queue is guaranteed to have space, so a
+        # simple non-blocking put is safe.  We still use put() (which
+        # blocks until space is available) as a defensive measure.
+        await queue.put(END_SENTINEL)
 
     async def subscribe(
         self,
@@ -84,7 +113,18 @@ class MemoryStreamBridge(StreamBridge):
             await asyncio.sleep(delay)
         self._queues.pop(run_id, None)
         self._counters.pop(run_id, None)
+        self._dropped_counts.pop(run_id, None)
 
     async def close(self) -> None:
         self._queues.clear()
         self._counters.clear()
+        self._dropped_counts.clear()
+
+    def dropped_count(self, run_id: str) -> int:
+        """Return the number of events dropped for *run_id*."""
+        return self._dropped_counts.get(run_id, 0)
+
+    @property
+    def dropped_total(self) -> int:
+        """Return the total number of events dropped across all runs."""
+        return sum(self._dropped_counts.values())