`ThrottlingRequestManager` never reopens its sub-queues, so a restart strands throttled requests

[janbuchar]

_get_or_create_sub_manager is only reached from add_request and add_requests, but every read
path — fetch_next_request, is_empty, is_finished, get_handled_count, get_total_count,
purge, drop — iterates _sub_managers. On a fresh process that dict is empty.

• Requests a previous run left in a throttled-<domain> queue stay invisible until something adds a
  new request for that same domain. Until then fetch_next_request takes its
  if not self._sub_managers: return None short-circuit and the crawl reports itself finished with
  work still queued.
• purge() and drop() skip them for the same reason, so they cannot be cleared either.

The sub-queues live under a stable alias specifically so they outlive the process, which this
defeats. The TS port opens every configured domain up front rather than on insert
(apify/crawlee#3741).

From #1762.