Actor & state-tree versioning

[Stebalien]

The VM itself needs to understand actor version for 3 reasons:

1. Actor lookup/registration in the init actor.
2. Actor lookup in the state-tree.
3. Network upgrades.

For 1, we could:

1. Understand all versions of the init actor.
2. Invoke methods on the current init actor wasm module to perform lookups/insertions.
3. Understand only the current version and version the entire FVM. To do that, we'd need a layer higher up the stack to multiplex between different FVM versions.

For 2, we could:

1. Understand all state-tree versions.
2. Compile the state-tree HAMT code to WASM, but that doesn't really buy us anything over (1).
3. Again, understand the current version and version the entire FVM.

For 3, we could:

1. Add the network upgrade logic to the FVM (implies understanding all versions).
2. Perform upgrades outside the FVM iteself (which I'd much prefer).

Finally: do we even support old actor versions in the FVM, or only new actor versions (6+)? We don't have v0 rust actors, and versions before v6 haven't been audited (and likely won't sync the chain).

Some interesting notes:

1. All we need is the init actor and the HAMT, nothing else.
2. The shape of the init actor has never changed, except for the HAMT itself.

[jennijuju]

The conclusion for now is that the implementation will handle the previous actor versions