fix(storage): propagate write errors instead of swallowing them in DatabaseCommit

[wpank]

Summary

• Short-circuit on commit failure: Moved the take_commit_failure() check from after the entire block execution loop to immediately after each per-transaction db.commit() call in revm.rs. This aborts block execution on the first QMDB write failure instead of continuing to execute remaining transactions against potentially inconsistent in-memory state.
• Enhanced error logging: Added up to 5 affected account addresses to the CRITICAL error log in the DatabaseCommit impl (adapter.rs) for easier diagnosis of write failures.
• Updated error message: Changed the log message to reflect that the executor now actively short-circuits on commit failure rather than continuing execution.

Problem

The REVM DatabaseCommit::commit() trait returns (), making error propagation impossible through the standard interface. When a per-transaction commit failed (e.g., disk I/O error), remaining transactions in the block continued executing against potentially stale or inconsistent in-memory state. The commit-failure flag was only checked after the entire block had finished, wasting computation and producing invalid receipts.

Test plan

• cargo check -p kora-executor -p kora-handlers passes
• cargo clippy -p kora-executor -p kora-handlers --no-deps -- -D warnings passes clean

Closes refcell#22

🤖 Generated with Claude Code