fix: RDP slot occupancy and cleanup_loop always running

- admin_page: slot shown as occupied based on ACTIVE status only (no time cutoff)
- go_service: busy slots checked by ACTIVE status (no cutoff) — cleanup_loop handles expiry
- startup_event: cleanup_loop starts regardless of ENABLE_STARTUP_MAINTENANCE flag;
  pool/container init guarded by the flag separately
- cleanup_loop: RDPSLOT sessions expire correctly and trigger container restart

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

app/main.py

@@ -1742,13 +1742,11 @@ def startup_event():
         fcntl.flock(lock_file.fileno(), fcntl.LOCK_UN)
     ensure_icons_dir()
     bootstrap_admin()
-    if not ENABLE_STARTUP_MAINTENANCE:
-        logger.info("startup_maintenance_disabled")
-        return
     if not try_acquire_maintenance_leader():
         logger.info("maintenance_leader_skipped")
         return
 
+    if ENABLE_STARTUP_MAINTENANCE:
         db = SessionLocal()
         try:
             ensure_universal_pool()
@@ -1765,8 +1763,14 @@ def startup_event():
                     slots = db.scalars(select(RdpSlot).where(RdpSlot.service_id == svc.id)).all()
                     for slot in slots:
                         try:
+                            cname = _rdp_slot_container_name(svc.slug, slot.id)
+                            try:
+                                c = docker_client().containers.get(cname)
+                                if c.status != "running":
+                                    c.start()
+                            except docker.errors.NotFound:
                                 start_rdp_slot_container(slot, svc)
-                        slot.container_name = _rdp_slot_container_name(svc.slug, slot.id)
+                                slot.container_name = cname
                         except Exception:
                             logger.exception("startup_rdp_slot_start_failed slot_id=%s", slot.id)
                     if slots:
@@ -1953,7 +1957,6 @@ def admin_page(request: Request, admin: User = Depends(require_admin), db: Sessi
         {"cutoff": cutoff},
     ).mappings().all()
     rdp_slots: dict[int, list] = {}
-    cutoff_slot = now_utc() - dt.timedelta(seconds=SESSION_IDLE_SECONDS)
     for svc in rdp_services:
         slots = db.scalars(select(RdpSlot).where(RdpSlot.service_id == svc.id).order_by(RdpSlot.id)).all()
         slot_list = []
@@ -1962,7 +1965,6 @@ def admin_page(request: Request, admin: User = Depends(require_admin), db: Sessi
                 select(SessionModel).where(
                     SessionModel.container_id == f"RDPSLOT:{slot.id}",
                     SessionModel.status == SessionStatus.ACTIVE,
-                    SessionModel.last_access_at >= cutoff_slot,
                 )
             )
             running = False
@@ -2162,12 +2164,10 @@ def go_service(
                     session_id = str(uuid.uuid4())
                     try:
                         with allocator_lock(db, 91003, timeout_seconds=GO_POOL_LOCK_TIMEOUT_SECONDS):
-                            cutoff = now_utc() - dt.timedelta(seconds=SESSION_IDLE_SECONDS)
                             busy_slot_ids: set[int] = set()
                             for row in db.scalars(
                                 select(SessionModel).where(
                                     SessionModel.status == SessionStatus.ACTIVE,
-                                    SessionModel.last_access_at >= cutoff,
                                     SessionModel.service_id == service.id,
                                     SessionModel.container_id.like("RDPSLOT:%"),
                                 )