Presently, the schema telemetry samples the catalog thusly:

• once a week, on midnight on the first day of the week
• if more than 10k records in a snapshot, picks 10k records at random

The log collection infrastructure does not have infinite capacity and would instead rather see a steady trickle of messages rather than an infrequent burst. We need to determine exactly what hard constraints we're operating under, and flatten the curve of the schema telemetry event logging peaks somehow. Either log less stuff, or for less clusters, or at more random times, or emit the logs more gradually, or all of the above.

Jira issue: CRDB-18432