Recover stalled database connections and skip unreachable chat recipients

tests/test_health.py

@@ -0,0 +1,31 @@
+import os
+from pathlib import Path
+import subprocess
+import sys
+
+
+def test_watchdog_exits_a_process_with_stalled_event_loop(tmp_path):
+    # A blocked event loop cannot run coroutine-based timeout handlers.
+    code = """import time
+from src.core.health import start_watchdog
+start_watchdog(grace_seconds=0.1, interval=0.02)
+time.sleep(10)
+"""
+    result = subprocess.run([sys.executable, "-c", code], cwd=Path(__file__).resolve().parents[1],
+                            env=dict(os.environ, HEARTBEAT_FILE=str(tmp_path / "absent")),
+                            capture_output=True, timeout=8)
+    assert result.returncode == 1
+    assert b"Heartbeat expired" in result.stderr
+
+
+def test_stopped_watchdog_allows_normal_shutdown(tmp_path):
+    code = """import time
+from src.core.health import start_watchdog
+stopped = start_watchdog(grace_seconds=0.1, interval=0.02)
+stopped.set()
+time.sleep(0.2)
+"""
+    result = subprocess.run([sys.executable, "-c", code], cwd=Path(__file__).resolve().parents[1],
+                            env=dict(os.environ, HEARTBEAT_FILE=str(tmp_path / "absent")),
+                            capture_output=True, timeout=8)
+    assert result.returncode == 0