kvserver: range descriptor passed to allocator should be used when calling change replicas

The range descriptor passed into `ChangeReplicas` is compared against the current descriptor within the `execChangeReplicasTxn`. If the descriptors are different, the replication change fails with a descriptor changed error, which is desirable given the ChangeReplicas was issued assuming a different set of replicas.

Currently, the replicate queue does not use the same descriptor from planning as it does for calling `ChangeReplicas`:

https://github.com/cockroachdb/cockroach/blob/1e8a4642d8fee46512f31514a53c6f22106c2b1b/pkg/kv/kvserver/replicate_queue.go#L814-L822

The store rebalancer has a slightly different variant of the same problem, because it calls `RelocateRange` instead of `ChangeReplicas` directly, so no descriptor is given:

https://github.com/cockroachdb/cockroach/blob/1e8a4642d8fee46512f31514a53c6f22106c2b1b/pkg/kv/kvserver/store_rebalancer.go#L731-L737

Note that validation does exist within `ChangeReplicas`, so that if the current descriptor is incompatible with the planned change, the change will fail:

https://github.com/cockroachdb/cockroach/blob/1e8a4642d8fee46512f31514a53c6f22106c2b1b/pkg/kv/kvserver/replica_command.go#L1831-L1849

This however doesn't check whether the range would become underreplicated or overreplicated from the proposed change.

Jira issue: CRDB-51105
Epic: CRDB-56265