Fix stale session causing persistent empty AP list

Two changes:

1. _request_with_retry: after a session-error re-login, also re-fetch
   sites and refresh the client reference so the retry uses valid state.

2. get_aps: if every site fails with an exception, clear the token and
   sites cache so the very next page reload triggers a fresh login.
   Previously the broken state was sticky until a container restart.

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

app/omada.py

@@ -103,9 +103,12 @@ class OmadaClient:
         resp.raise_for_status()
         data = resp.json()
         if data.get("errorCode", 0) in (-1006, -1003, -30109):
+            logger.info("Session error %s — re-authenticating", data.get("errorCode"))
             self._token = None
             self._sites = None
             await self._login()
+            await self._fetch_sites()
+            client = self._get_client()
             resp = await client.request(method, url, headers=self._auth_headers(), **kwargs)
             resp.raise_for_status()
             data = resp.json()
@@ -160,6 +163,8 @@ class OmadaClient:
         """Fetch APs from every accessible site and return a combined list."""
         await self._ensure_ready()
         all_aps = []
+        failed_sites = 0
+        total_sites = len(self._sites)
         for site_name, site_key in self._sites.items():
             try:
                 data = await self._request_with_retry(
@@ -174,7 +179,15 @@ class OmadaClient:
                         all_aps.append(d)
             except Exception as exc:
                 logger.warning("Failed to fetch devices for site '%s': %s", site_name, exc)
-        logger.info("Fetched %d APs across %d sites", len(all_aps), len(self._sites))
+                failed_sites += 1
+        if failed_sites > 0 and failed_sites == total_sites:
+            # Every site failed — session is likely stale with an unrecognised error code.
+            # Clear auth state so the next request triggers a full re-login instead of
+            # requiring a container restart.
+            logger.warning("All %d site(s) failed — clearing session state to force re-auth on next request", total_sites)
+            self._token = None
+            self._sites = None
+        logger.info("Fetched %d APs across %d sites", len(all_aps), total_sites)
         return all_aps
 
     async def _get_ap_uptime(self, mac: str, site_key: str) -> int: