A tracing span should expose a listener API that can be used to listen in on structured events that are recorded by the span and its children. This is motivated by the need for higher-level aggregators such as #80388 to be able to aggregate information from interesting events that are recorded throughout a root spans lifetime. An event listener allows us to be certain that we won't miss events that are rotated out of the underlying ring buffer due to memory constraints. Jobs for example can run for hours and emit thousands of structured events, we want to be able to account for every event in our metrics to power our observability tools accurately.

Epic: CRDB-10262

Jira issue: CRDB-15626