fix: check hashed state for loading TriePrefixSets::destroyed_accounts

[joshieDo]

During a pipeline unwind, the MerkleUnwind stage will deal with an unwound hashed state and a non-unwound plain state. On normal execution, destroyed_accounts are populated if accounts are no longer present on PlainState. This will eventually delete any existing storage tries.

However, on a unwind to a block where an address is no longer present and used to have storage tries, this does not happen, since we only unwind PlainState after the MerkleUnwind. Resulting in forgotten storage tries. Leading to the the following failure case on holesky:
2631791 → unwind 2631774 ✅ → 2631775  state root error ❌

By looking at hashed state instead, this can be solved

edit:

Performance wise I don't know if this has an impact actually. We're searching a table with 32byte key instead of 20byte. Should additional logic be added? if _unwind_ { check hashed_state } else { check plain_state }

[mattsse]

pending @rkrasiuk

[rkrasiuk]

@joshieDo good find, lgtm. wouldn't worry about performance too much here since pipeline unwinds are supposed to be rare