fix: use current exec state when querying validator table

[karlem]

Close #1233

The issue has been resolved by using the current execution state instead of the previous state retrieved through the read-only view. This ensures that the validators cache accurately reflects the latest set of validators, preventing the panic.

[cryptoAtwill]

@karlem Somehow the first impression that strikes is shouldnt we try to update the read_only_view to reflect latest state instead of passing mutable ref of state to a read only function?

[raulk]

@cryptoAtwill Unfortunately the FVM executor does not expose read-only semantics through its execute_message API.

So we'd have to add a method to "fork" the executor/machine using the intermediate state tree root (which you can get from machine.state_tree()) and a BufferedBlockstore over the original Blockstore (machine.blockstore()), which we discard after we finish the query.

Alternatively, we could forego this state query entirely and:

1. Guarantee we have always have a fully initialized ValidatorCache at the start.
2. Merge the power updates onto our ValidatorCache.

[raulk]

Merging this and opening #1236 to fix the right way.