When #98433 lands, we should consider limiting the number of concurrent lease requests in flight, since we'll be extending 1/6 of all local leases every 500 ms, probably by using a semaphore during requestLease(). However, prioritization can be a bit tricky here -- for example, we probably want lease acquisitions that's blocking foreground traffic to take precedence over lease extensions for mostly-idle ranges.

We can also consider having a bounded goroutine pool for lease requests, instead of spawning new goroutines for these.

Jira issue: CRDB-26429

Epic CRDB-25207