admission: intra-tenant inter-{database,workload} performance isolation

[irfansharif]

Is your feature request related to a problem? Please describe.

AC tries to provide performance isolation across tenants by maintaining per-tenant work in individual work queues and prioritization of work only within a work queue. There's nothing special about the tenant boundary as far as AC is concerned; the tenant ID is treated opaquely as the thing we partition work queues by. This issue tracks prototyping a version of this where instead we segment by an ID derived from in-tenant SQL state, like the database/workload/application name. It's common in SaaS applications to effectively implement multi-tenant along database boundaries where we'd want to provide performance isolation between databases/workloads. Likely depends on fixing #85471.

Additional context

#89721 is a test we’re taking over/nursing back to health. We also expect to add many more variants of it with mixed-workloads across mixed tenants (as part of #89208). We mention it because it’ll teach us (or me at least) how effective we are today at providing performance isolation across tenant boundaries. If we were to do intra-tenant performance isolation like described above, these learnings will apply as well.

Jira issue: CRDB-20681