sql: track self-induced latch wait time as a separate metric

We already have a `contention time` metric which is visible in statement bundles and statement statistics. This includes time spent waiting on another transaction in the lock table and (after https://github.com/cockroachdb/cockroach/pull/113649) in the latch table. However, the latch table is also used to wait on requests issued by the _current_ transaction. QueryIntent requests are an example of this; they are used to wait until a pipelined write has finished replicating. Since the cause and mitigation for high self-induced latch wait time is different from contention time, it is not included in the `contention time` metric.

This issue tracks adding a separate metric for self-induced latch wait time. Note that after https://github.com/cockroachdb/cockroach/pull/113649 merges, self-induced latch wait events will be visible in the trace as lines like `event:latch table conflict with 285e526b-1cbe-41d7-982b-bb043762410a on /Table/106/1/0/0 for 0.909s`.

Jira issue: CRDB-38684

Start by tracing the existing contention time metric through statement bundles and statement statistics, using the described latch-table trace events and QueryIntent requests as the behavioral reference. Define completion as a separate metric for self-induced latch wait time that is not included in contention time, and verify its visibility in the relevant statistics surfaces.

Written by the indexing model from the issue text.