fix(retention): add cleanup sweeps + CalDAV orphan reconciliation

Drift-audit Group 4 (retention / unbounded growth):

- CalDAV pull now reconciles deletions: a previously-synced event whose
  caldav_uid no longer appears remotely within the synced window is
  soft-deleted (one batch_id per run, restorable), so a remote delete
  propagates locally instead of orphaning forever. Guarded on a non-empty
  fetch so a spurious empty result can't wipe every local copy. Also wrap
  the blocking fetch in a 120s wait_for and log run duration.
- Notifications: hourly loop now purges read notifications older than 30d
  (unread kept). Table no longer grows without bound.
- Auth tokens: new daily sweep deletes password-reset / invitation tokens
  whose validity window ended >7d ago; wired via start_auth_token_retention_loop
  in app startup. Both tables previously only flipped used=True, never pruned.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/fabledassistant/app.py

@@ -148,12 +148,14 @@ def create_app() -> Quart:
     async def startup():
         import asyncio
 
+        from fabledassistant.services.auth import start_auth_token_retention_loop
         from fabledassistant.services.embeddings import backfill_note_embeddings
         from fabledassistant.services.logging import start_log_retention_loop
         from fabledassistant.services.notifications import start_notification_loop
 
         start_log_retention_loop()
         start_notification_loop()
+        start_auth_token_retention_loop()
 
         # Backfill embeddings for any notes that don't have one. Runs in the
         # background so it never blocks the server from accepting requests.

src/fabledassistant/services/auth.py

@@ -356,3 +356,46 @@ async def revoke_invitation(invitation_id: int) -> bool:
         await session.commit()
         logger.info("Invitation %d revoked", invitation_id)
         return True
+
+
+# ── Token retention ─────────────────────────────────────────────────────
+# Password-reset and invitation tokens are only ever flipped used=True and are
+# never pruned, so on a long-lived instance both tables grow without bound.
+# A daily sweep deletes any token whose validity window ended over grace_days
+# ago (covers both used and naturally-expired rows once they're cold).
+
+_auth_retention_task = None
+
+
+async def purge_expired_auth_tokens(grace_days: int = 7) -> int:
+    """Delete password-reset / invitation tokens that expired > grace_days ago."""
+    from sqlalchemy import delete
+    cutoff = datetime.now(timezone.utc) - timedelta(days=grace_days)
+    removed = 0
+    async with async_session() as session:
+        for model in (PasswordResetToken, InvitationToken):
+            result = await session.execute(
+                delete(model).where(model.expires_at < cutoff)
+            )
+            removed += result.rowcount or 0
+        await session.commit()
+    return removed
+
+
+async def _auth_token_retention_loop() -> None:
+    import asyncio
+    while True:
+        await asyncio.sleep(86400)  # daily
+        try:
+            removed = await purge_expired_auth_tokens()
+            if removed:
+                logger.info("Auth token retention: deleted %d expired token(s)", removed)
+        except Exception:
+            logger.exception("Error in auth token retention cleanup")
+
+
+def start_auth_token_retention_loop() -> None:
+    global _auth_retention_task
+    import asyncio
+    if _auth_retention_task is None or _auth_retention_task.done():
+        _auth_retention_task = asyncio.create_task(_auth_token_retention_loop())

src/fabledassistant/services/caldav_sync.py

@@ -11,7 +11,7 @@ import uuid
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
-from sqlalchemy import select
+from sqlalchemy import select, update
 
 from fabledassistant.models import async_session
 from fabledassistant.models.event import Event
@@ -20,6 +20,9 @@ logger = logging.getLogger(__name__)
 
 _SYNC_PAST_DAYS = 30
 _SYNC_FUTURE_DAYS = 180
+# Wall-clock cap on the blocking CalDAV fetch so a hung/slow server can't
+# wedge the hourly sweep indefinitely.
+_SYNC_TIMEOUT_SECONDS = 120
 
 
 def _parse_dt(val: Any) -> datetime | None:
@@ -116,16 +119,24 @@ async def sync_user_events(user_id: int) -> dict:
 
     config = await get_caldav_config(user_id)
 
+    started = datetime.now(timezone.utc)
+    range_start = started - timedelta(days=_SYNC_PAST_DAYS)
+    range_end = started + timedelta(days=_SYNC_FUTURE_DAYS)
+
     loop = asyncio.get_running_loop()
     try:
-        remote_events: list[dict] = await loop.run_in_executor(
-            None, _sync_one_user, config, user_id
+        remote_events: list[dict] = await asyncio.wait_for(
+            loop.run_in_executor(None, _sync_one_user, config, user_id),
+            timeout=_SYNC_TIMEOUT_SECONDS,
         )
+    except asyncio.TimeoutError:
+        logger.warning("CalDAV pull sync timed out for user %d after %ds", user_id, _SYNC_TIMEOUT_SECONDS)
+        return {"error": "CalDAV fetch timed out"}
     except Exception:
         logger.warning("CalDAV pull sync failed for user %d", user_id, exc_info=True)
         return {"error": "CalDAV fetch failed"}
 
-    created = updated = unchanged = skipped = 0
+    created = updated = unchanged = skipped = deleted = 0
 
     async with async_session() as session:
         for ev in remote_events:
@@ -185,13 +196,38 @@ async def sync_user_events(user_id: int) -> dict:
                 else:
                     unchanged += 1
 
+        # Reconcile deletions: a previously-synced event (has a caldav_uid)
+        # that no longer appears remotely within the synced window is
+        # soft-deleted, so a delete on the remote propagates locally instead
+        # of orphaning forever. Guarded on a non-empty fetch so a spurious
+        # empty result can't wipe every local copy.
+        if remote_events:
+            remote_uids = {e["caldav_uid"] for e in remote_events}
+            orphan_batch = str(uuid.uuid4())
+            orphan_res = await session.execute(
+                update(Event)
+                .where(
+                    Event.user_id == user_id,
+                    Event.caldav_uid.isnot(None),
+                    Event.caldav_uid.notin_(remote_uids),
+                    Event.deleted_at.is_(None),
+                    Event.start_dt >= range_start,
+                    Event.start_dt <= range_end,
+                )
+                .values(deleted_at=datetime.now(timezone.utc), deleted_batch_id=orphan_batch)
+            )
+            deleted = orphan_res.rowcount or 0
+
         await session.commit()
 
+    elapsed = (datetime.now(timezone.utc) - started).total_seconds()
     logger.info(
-        "CalDAV sync user %d: %d created, %d updated, %d unchanged, %d skipped (trashed)",
-        user_id, created, updated, unchanged, skipped,
+        "CalDAV sync user %d: %d created, %d updated, %d unchanged, %d skipped (trashed), "
+        "%d deleted (orphaned) in %.1fs",
+        user_id, created, updated, unchanged, skipped, deleted, elapsed,
     )
-    return {"created": created, "updated": updated, "unchanged": unchanged, "skipped": skipped}
+    return {"created": created, "updated": updated, "unchanged": unchanged,
+            "skipped": skipped, "deleted": deleted}
 
 
 async def sync_all_users() -> None:

src/fabledassistant/services/notifications.py

@@ -236,6 +236,28 @@ async def check_due_tasks() -> None:
                 logger.exception("Failed to send task reminder for user %d", user_id)
 
 
+# Read notifications are kept this long before the hourly sweep deletes them;
+# unread are kept regardless. Without a sweep the table grows without bound.
+_NOTIFICATION_RETENTION_DAYS = 30
+
+
+async def purge_old_read_notifications(retention_days: int = _NOTIFICATION_RETENTION_DAYS) -> int:
+    """Delete already-read in-app notifications older than retention_days."""
+    from datetime import timedelta
+    from sqlalchemy import delete
+    from fabledassistant.models.notification import Notification
+    cutoff = datetime.now(timezone.utc) - timedelta(days=retention_days)
+    async with async_session() as session:
+        result = await session.execute(
+            delete(Notification).where(
+                Notification.read_at.isnot(None),
+                Notification.read_at < cutoff,
+            )
+        )
+        await session.commit()
+        return result.rowcount or 0
+
+
 async def _notification_loop() -> None:
     while True:
         await asyncio.sleep(3600)  # hourly
@@ -243,6 +265,12 @@ async def _notification_loop() -> None:
             await check_due_tasks()
         except Exception:
             logger.exception("Error in notification loop")
+        try:
+            removed = await purge_old_read_notifications()
+            if removed:
+                logger.info("Notification retention: deleted %d read notification(s)", removed)
+        except Exception:
+            logger.exception("Error in notification retention cleanup")
 
 
 def start_notification_loop() -> None: