kvserver: reduce reproposals of AddSSTable and other large commands

We've seen that Raft can run nodes out of memory during overload, in particular with large commands like `AddSSTable` (see e.g. #73376). One possible contributing factor here is reproposals, where an overloaded Raft group that takes more than 3 seconds to ack a proposal may continually stuff the same command into Raft over and over, worsening the overload. We should consider throttling reproposals of large commands.

Jira issue: CRDB-25334

Start by reviewing the overload and reproposal context in #73376, then follow the linked Jira issue CRDB-25334 for the intended scope. Investigate how large commands such as AddSSTable are reproposed during delayed acknowledgements, and define the throttling behavior and overload-memory evidence needed to consider the issue done.

Written by the indexing model from the issue text.