If a job is paused, it retains claim_session_id and claim_instance_id columns.
Thus, if it is subsequently resumed, the same node will reclaim the job...
We should clear out both columns when pausing the job.

Currently, this is the only mechanism we have to try to "rebalance" jobs resumer
assignment.

Also consider lowering the number of jobs registry can claim at a time.
Bonus: consider moving this limit from an env var to a setting.

Jira issue: CRDB-16553

Epic CRDB-15034