debug(backup): 5s settle + INFO log on entity-not-yet-indexed

Two diagnostics for the persistent WS-backed-domain capture failures
across e2e (helper, label, category, zone, area, group,
dashboard_resource):

1. Bump test create→edit settle to 5 s. 2 s wasn't enough on CI under
   load — every WS-list capture was returning None silently and the
   1288 full-loop tests timed out at 15 s waiting for a snapshot that
   never came.

2. Add an INFO log when ``maybe_snapshot``'s fetch returns None so the
   "entity not yet in WS-list" case is visible in test logs. Previously
   silent — the only WARN paths were transient errors or write
   failures, neither of which fired here. Marked for downgrade to DEBUG
   once the full-loop e2e is stable; surfacing it at INFO short-term so
   the diagnostic crumb is in CI for the next iteration.

If 5 s settle still produces "fetch returned None" logs, the issue
isn't propagation latency — it's that the create's response shape
differs from the list shape (id mismatch). The INFO line will tell.

Author: kingpanther13

src/ha_mcp/backup_manager.py

@@ -278,6 +278,14 @@ async def maybe_snapshot(
             if config is None:
                 # Entity didn't exist at fetch time (create operation, or
                 # already-deleted at remove time before our pre-fetch).
+                # Logged at INFO so e2e diagnostics surface the
+                # "entity-not-yet-indexed" case without polluting prod
+                # WARNING streams; downgrade to DEBUG later once #1288
+                # full-loop e2e coverage is stable.
+                logger.info(
+                    "Auto-backup: fetch returned None for %s — skipping snapshot",
+                    key,
+                )
                 return None
             try:
                 path = await asyncio.to_thread(

tests/src/e2e/workflows/auto_backup/test_capture_and_restore.py

@@ -55,11 +55,10 @@ def _backups_for(
 # fetch fires. The decorator is best-effort: if fetch returns None
 # (entity not in the registry list yet), the snapshot is silently
 # skipped — and polling the backup file afterwards can't recover that.
-# 2 s is the empirically-stable mark on HAOS testcontainer runners;
-# automation's REST upsert path settles faster and doesn't need this,
-# but every WS-backed domain (label, category, zone, area, helper,
-# dashboard_resource) does.
-_HA_PROPAGATION_SETTLE_SECONDS = 2.0
+# 5 s is conservative on CI runners under load; automation's REST
+# upsert path settles faster and doesn't need this, but every WS-backed
+# domain (label, category, zone, area, helper, dashboard_resource) does.
+_HA_PROPAGATION_SETTLE_SECONDS = 5.0
 
 
 async def _wait_for_backup(