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

The rttanalysis framework is helpful because it lets us track round-trips used to run certain queries. Round-trips are not everything. While in multi-region settings, round-trips often dominate the cost of execution, deserializing and validating many descriptors is more expensive. In some cases we can do an index-join to look up a small number of descriptors and that's preferable to scanning all descriptors if the number of descriptors is large (#90116).

Describe the solution you'd like
We should also track the number of descriptors which need to be decoded and other metrics related to the cost of an operation.

Jira issue: CRDB-21043