Improve `sameDomainDelaySecs` implementation

[janbuchar]

• The current implementation (added in feat: add support for sameDomainDelay #2003) essentially holds requests to recently accessed domains in memory and re-enqueues them after a given timeout
• ⚠️ this may result in lots of unnecessary request queue writes
• ⚠️ there may be unexpected results if this is used in conjunction with request locking
  • but resolving this for multiple request queue consumers would be quite an endeavor
• we cannot simply use maxRequestsPerMinute - we don't know the request URL (and thus the domain) before receiving it from the queue

Regardless of the resolution, the "final" version should be ported over to crawlee-python.

[janbuchar]

I believe that we could use a separate unnamed request queue for each domain that should be delayed. The queue could simply keep track of the delay and only return a new request after it passes.

This would only be efficient if the sorting of requests into queues was done immediately - popping requests from the "default" request queue and then re-enqueueing them into specific, delayed queues would cost a lot of RQ operations.