rangefeed: user-defined quality-of-service prioritization

[erikgrinaker]

Different changefeeds on the same node may have different requirements: some may need low latency, while others are not latency-sensitive at all. It would be nice if users could specify the relative priority between rangefeeds in case of resource contention. This would involve:

• Scheduler priority.
• Mux stream priority.
• Memory budget priority.
• gRPC/TCP connection priority.

It would probably also need higher-level prioritization of the writing workload, e.g. in AC and other systems.

Jira issue: CRDB-31509

Epic CRDB-39959

[blathers-crl[bot]]

cc @cockroachdb/replication

[erikgrinaker]

cc @miretskiy in case you have something to add.