It would be useful to report on inter-statement latency within transactions. That is, how long do we spend waiting for the SQL client to send us a new statement after the result-set of the previous statement has drained?

We see cases in which a "slow" transaction spends the majority of its time idle, waiting for the caller to, e.g., call some other micro-service, which leads to contention, etc. If we can identify workloads that have high inter-statement latency, we can recommend to customers some places where they would get the biggest return on breaking a single, long-running, mostly-idle transaction into multiple transactions.

Jira issue: CRDB-18865

Epic CRDB-20388