Describe the problem

This has huge spillover impacts on the kvsubscriber in the spanconfig package because it means that it is not guaranteed to ever succeed.

To Reproduce

Create over 1000 tables. Watch it fail.

Expected behavior
It succeeds.

Additional context

Terrifyingly, on this cluster I saw ranges merge down to just 1. That shouldn't happen.

We should definitely test spanconfig reconciliation performance on large schemas.

Jira issue: CRDB-13712