Fix early reminder notifications and lost recovery notifications

- AlertState.update() now resets last_notification when the alert level
  changes, so a WARNING→CRITICAL escalation restarts the reminder interval
  rather than inheriting a nearly-expired timer.
- _dispatch_to_channel() bypasses min_level for RECOVER, so recovery
  notifications are delivered even after a server restart when
  _alerted_channels is empty and the fallback dispatch path is used.

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

hbd/server/notify.py

@@ -385,13 +385,20 @@ _DRIVERS = {
 
 
 def _dispatch_to_channel(channel_name: str, channel_cfg: dict, notif: Notification) -> bool:
-    """Send *notif* to a single named channel, honouring min_level."""
-    min_level = channel_cfg.get("min_level", "WARNING").upper()
-    if _level_value(notif.level) < _level_value(min_level):
-        logger.debug(
-            "channel '%s': skipping level %s (min_level=%s)", channel_name, notif.level, min_level
-        )
-        return True  # not an error — filtered intentionally
+    """Send *notif* to a single named channel, honouring min_level.
+
+    RECOVER always bypasses min_level — a recovery is always relevant if the
+    channel was configured for any alerting (handles the restart-then-recover case
+    where _alerted_channels is empty and we fall through to the normal loop).
+    """
+    level = notif.level.upper()
+    if level != "RECOVER":
+        min_level = channel_cfg.get("min_level", "WARNING").upper()
+        if _level_value(level) < _level_value(min_level):
+            logger.debug(
+                "channel '%s': skipping level %s (min_level=%s)", channel_name, level, min_level
+            )
+            return True  # not an error — filtered intentionally
 
     ch_type = channel_cfg.get("type", "")
     driver = _DRIVERS.get(ch_type)