admission: support for network IO prioritization

This support ticket was created because TTL caused a client to reached the network IO limit of their cluster:
https://github.com/cockroachlabs/support/issues/2612

I think we need something like admission control to naturally rate limit the TTL job by prioritizing its network traffic.

The current solution is to set `ttl_delete_rate_limit` which is difficult for the customer to configure and changes with the number of table range leaseholders.

This is similar to https://github.com/cockroachdb/cockroach/issues/107623, but for network IO bandwidth instead of storage IO bandwidth.

Jira issue: CRDB-31756