migrations part 2/2: persistence

This finishes up the bulk of the remaining work for the [version migation
RFC][rfc]. There are two main changes:

1. Add `*ExposedClusterVersion` to the `cluster.Settings` struct. Internally,
   it wraps the `version` cluster setting, but it exposes it indirectly to
   synchronize version updates with the node's engines. More precisely,
   - `*ExposedClusterVersion` is initialized with the persisted version early
     in the `Node` startup sequence
   - the `Node` registers itself for notifications whenever the `version`
     changes
   - `*ExposedClusterVersion` only exposes a change in the underlying `version`
     variable to callers *after* calling the `Node` callback (which in turn
     persists the update to all engines).
   - The result is that whenever `IsActive()` or `Version()` is called, the
     returned version is safe to use or to migrate to without having to worry
     about backwards compatibility.
1. Add code that synthesizes from and writes the version to the engines.
   This mostly picks up #17155 with some code golf that refactors the node
   startup sequence appropriately, making sure that even when there are stores
   that need (asynchronous) bootstrapping, they'll still have the cluster
   version persisted at the end.

TODO:

- [ ] Testing.
- [ ] Update the RFC.
- [ ] Pave the way for future use of `UseVersion < MinVersion`, if still required.
- [ ] Migration that makes sure there's always a cluster version settings able
entry (@m-schneider).
- [ ] Disallow version bump if settings table has no value entry.
- [ ] Consider removing the reliance on gossip callback ordering. Gossip does
      not *guarantee* that version updates arrive in-order, and we currently
      persist the cluster version blindly. `*ExposedClusterVersion` would
      have to do its own synthesis to refuse illegal changes backwards. This
      may be a non-issue but needs some consideration (all of the cluster
      settings would be vulnerable to it).

[rfc]: cockroachdb/cockroach#16977

v3_test.go

@@ -38,7 +38,7 @@ func makeTestV3Conn(c net.Conn) v3Conn {
 	mon := mon.MakeUnlimitedMonitor(
 		context.Background(), "test", mon.MemoryResource, nil, nil, 1000,
 	)
-	st := cluster.MakeClusterSettings()
+	st := cluster.MakeTestingClusterSettings()
 	exec := sql.NewExecutor(
 		sql.ExecutorConfig{
 			AmbientCtx:              log.AmbientContext{Tracer: st.Tracer},