Add metrics for sql service latency broken down by statement type

Today we have SQL service latency metric (called `sql_service_latency_bucket` in prometheus) for overall SQL traffic. Some customers find it valuable to break this down by SELECT, INSERT, DELETE, UPDATE etc so they can understand performance characteristics better. To quote a customer:
>The idea is that update, insert, select, and to some extent delete have different > performance characteristics and resource needs, so it might be useful to have e.g. >cr.node.sql.insert_service.latency-p50, cr.node.sql.update_service.latency-p50, >cr.node.sql.select_service.latency-p50 instead of just cr.node.sql.service.latency-p50

Jira issue: CRDB-35777