rangeDesc: provide an RDI iterator integrated with elastic admission control

[msbutler]

We observed in #147759 that backup's use of an RDI can degrade the foreground workload. We suspect that integrating elastic admission control into the RDI could prevent this.

Additional slack thread here.

Jira issue: CRDB-51626

[blathers-crl[bot]]

Hi @msbutler, please add branch-* labels to identify which branch(es) this C-bug affects.

_{:owl: Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.}

[dshjoshi]

@msbutler Is this work that is in backup area of in KV? And what is the priority of this ticket?

[dshjoshi]

@msbutler @arulajmani Will this be part of KV work or DR work? - Thanks!

[arulajmani]

@msbutler any reason this was marked as a P-1? IIUC, this issue is very similar to other issues where foreground latency is impacted because of O(nodes/stores/ranges) work, all of which are marked as P-3s (e.g.https://github.com/cockroachdb/cockroach/issues/148211)

@dshjoshi this falls into KV.

[msbutler]

@arulajmani no strong reason this was marked as a P1. Feel free to reprioritize.

[msbutler]

Action item: grab a cpu profile at the beginning of a backup

[dshjoshi]

@arulajmani @msbutler Given that , this is marked as P-3 ; can we remove "300-node-qualification-blocker" label from it?

[dshjoshi]

Confirmation from Rima that this is not "300-node-qualification-blocker"