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

This issue tracks deferred work from PRs that introduced replication admission control (#95563, #98308). This list is sourced partly using checked-in TODOs ($ rg "#95563" --before-context 4).

• kvserver: remove above-raft throttling for AddSST #102683
• Export metric to track total AddSST bandwidth, both proposed and applied (this was prototyped in #93102, addsstable.{proposal,application}_bytes)
  • The aggregated elastic flow tokens deducted metric works well as a substitute
• kvflowcontroller: mutex contention #105508
  • TODO: Eliminate heap allocation in kvflowcontrol.ContextWithMeta
  • TODO: Write microbenchmark for allocation profile of WorkQueue.Admit, and optionally improve it.
• TODO: Eliminate OOM possibility when kvadmission.flow_control.mode = apply_to_elastic.
• TODO: Add a byte/count limit to how many flow token dispatches can be added to a raft message
• Surface some kvflowcontrol metrics in the admin UI.
• Address remaining performance cuts identified in kvflowcontrol: annotate/fix perf regressions #109833.
• Dump vtable state in debug.zip
• Don't use flow tokens (or AC really) for lease requests (see this comment).
• Don't use flow tokens or AC when writing to system tables (especially around writes to cluster settings)
• TODO: Fix over-admission due to AC token acquisition fast path.
• TODO: Remove O(local stores) lookup for flow control handles.

Epic CRDB-25348

Jira issue: CRDB-28374

Epic CRDB-25348