feat(scanner): add work queue for book dock pipeline and auto-size node heap

Introduce BookDockWorkQueue to serialize metadata extraction and auto-finalize
jobs. Files process newest-first with a 250 ms drain window and 500-1000 ms
inter-job spacing to avoid hammering metadata providers. Watched-folder rescans
now requeue files stuck in extracting or fetching status after a server restart
rather than silently dropping them.

Replace the global ThrottlerGuard with SensitiveEndpointThrottlerGuard, which
only activates on endpoints that carry explicit @Throttle() metadata. This
eliminates all @SkipThrottle() decorators from non-auth controllers and adds
missing throttle coverage for oidcGenerateLinkState and oidcGeneratePreviewState.
Update the login page to match the simplified 429 error message.

Fix countsByStatus to aggregate extracting and fetching rows into the pending
bucket so the UI counter stays accurate mid-pipeline. Add a secondary id sort
to findAll for deterministic pagination ordering.

Auto-detect the container memory limit via cgroup v1/v2 in entrypoint.sh and
compute --max-old-space-size proportionally (50-75 % of limit) instead of
defaulting to a fixed 2048 MB that OOMs small containers.

Closes #441

docker-compose.yml

@@ -25,7 +25,7 @@ services:
       CLIENT_URL: ${CLIENT_URL:-${APP_URL?required}}
       PUID: ${PUID:-1000}
       PGID: ${PGID:-1000}
-      NODE_MAX_OLD_SPACE_SIZE: ${NODE_MAX_OLD_SPACE_SIZE:-2048}
+      NODE_MAX_OLD_SPACE_SIZE: ${NODE_MAX_OLD_SPACE_SIZE:-auto}
       BOOK_DOCK_PATH: ${BOOK_DOCK_PATH:-}
     depends_on:
       postgres: