[vm] Delta writes in sequential execution

[georgemitenkov]

Description

Slightly refactored delta operations so that Aggregator 1) can reuse apply_to functionality based on operation 2) can reuse unique serialization/deserialization for delta op values (This way it should also be easier to extract these deltas to ChangeSet, etc.). Added tests.

Implemented delta application for sequential execution. @gelash @zekun000 I am looking for a place to put tests in the form of a) generate writeset with deltas b) apply with fake executor / fake data store c) check result of application and/or VM status. I suppose e2e-tests/src/executor.rs might be the best place for that? Or a separate file?

Test Plan

• Unit tests
• Fake executor tests

---

This change is 

[davidiw]

overall seems fine...

My bigger concern is on the first PR, where we use unreachable instead of error, where we have yet to implement the validation to make sure that a Delta::Op can't leave the VM / executor.

[georgemitenkov]

@davidiw Speaking about unreachable: there are 3 places where this is the case:

1. transaction replay: we match by AccessPath and since we don't use it for aggregator's state key - should never be exposed.
2. in try_access_path_into_write_set_change and try_table_item_into_write_set_change - again, AccessPath is never used for aggregator.
3. In InMemoryStateCalculator: this is used for storage and deltas should be converted by then.

I agree that this is not nice but at least it allows to proceed with implementation on executor side (and we not cresting deltas just yet anywhere, so fine for now?). I am working on moving this to ChangeSet or TransactionOutput (separate delta set) in parallel.

[georgemitenkov]

Will be rebasing this on #2222.

TODOs: prologue/epilogue