fix(database): allow EIP161 state clear for empty Loaded and Changed accounts

[msheth-circle]

Summary

• Fix on_touched_empty_post_eip161() panic when called on a Loaded AccountStatus
• Allow Loaded and Changed → Destroyed transition unconditionally

On EVM chains where the native token is an ERC20, an EOA can hold the native token (balance > 0, nonce=0, no code) and spend it all as an ERC20 via a meta-transaction (relayer pays gas, no nonce change on the EOA), making the account empty. In this scenario, the account was loaded from DB with Loaded status, and since the balance change is tracked in the Journal — not through CacheAccount::change() — the CacheAccount status remains Loaded when touch_empty_eip161 is called.

The fix allows Loaded → Destroyed unconditionally. This is safe because touch_empty_eip161 is only reachable when account.is_empty() is true on the Journal output, which requires no code (code_hash == KECCAK_EMPTY). A contract in Loaded status cannot reach this path — SELFDESTRUCT is the only way to remove code, and it's handled by the is_selfdestructed() early return before this point. This is consistent with geth, which has no status distinction and simply deletes any dirty empty account in Finalise().

Test plan

• Added unit test for Loaded → Destroyed transition in on_touched_empty_post_eip161
• Added #[should_panic] test confirming Changed still panics
• cargo test -p revm-database — 31 passed
• cargo nextest run --workspace — 369 passed, 0 failed
• cargo clippy --workspace --all-targets --all-features — 0 warnings

[codspeed-hq]

Merging this PR will not alter performance

✅ 176 untouched benchmarks

---

_{Comparing msheth-circle:fix/eip161-loaded-account (17e4f0f) with main (3bccc97)}

[msheth-circle]

@rakita Mind taking another look?

[rakita]

Merging after CI

[msheth-circle]

Thanks @rakita! CI is looking good

[rakita]

Will make release soon, in next few days