renja-g/RiftRelay

AI Summary: This issue proposes the implementation of "Unbound Priority Levels" to categorize requests with a flexible, unbounded scale. Key considerations include how these priorities interact with pacing mechanisms, whether to implement strict priority or a fair queuing algorithm like WFQ, and the potential performance implications.

AI Summary: The RiftRelay service currently allows multiple parallel requests to pass through without rate limiting when it's first started and hasn't learned any limits. This can lead to exceeding API rate limits, as demonstrated by the example using `parallel` and `curl`. A potential solution is to implement default rate limits, such as those for development keys, to govern requests during the initial learning phase.