`ThrottlingRequestManager.fetch_next_request` holds a concurrency slot while waiting out a throttle

[janbuchar]

• When every configured domain is throttled and the inner manager is empty, fetch_next_request sleeps for up to max_delay rather than returning. It's called from __run_task_function, so that's an autoscaled-pool task slot held open doing nothing — which the snapshotter is liable to read as spare capacity and answer by scaling up.
• The TS port returns None and has is_empty() report True so the task loop idles instead (feat: implement per-domain request throttling (ThrottlingRequestManager) crawlee#3741). Costs a contract bend — is_empty() is True while requests still exist — but it stops lying to the autoscaler.
• From feat: Add opt-in per-domain request throttling for HTTP 429 backoff #1762.