As discovered in #111561, the IsActive cluster version check is on many hot paths. Today it incurs a protobuf Unmarshal, which until recently also allocated (#113043) and amounted as much as 1% of allocations in some tests.

While we got rid of the allocation, we should take it further and be able to do the cluster version checks without unmarshaling protos in the first place.

@dt says:

we could also just move clusterversion.proto into another package that pkg/settings is allowed to depend on and then store a *ClusterVersion in the atomic instead of a []byte so there'd be no unmarshal -- allocating or otherwise -- on the hot path.

Jira issue: CRDB-32919