Harden lmdblib LMDB failure handling

[AztecBot]

Summary

lmdblib treated several LMDB failure statuses as normal not-found or end-of-data cases, and transaction failure paths could leave guard or transaction state inconsistent. This change preserves the existing not-found behavior while throwing on real LMDB errors and making transaction lifecycle cleanup deterministic.

Fix

• Distinguish MDB_NOTFOUND from real LMDB errors in query and cursor helpers.
• Make stats reads use initialized output and the throwing mdb_stat wrapper.
• Mark failed commits terminal before throwing and add RAII cleanup for reader/writer guard acquisition.
• Add focused lmdblib regressions for invalid duplicate cursor operations, empty filtered previous lookup, uncommitted stats, and committed transaction reuse.

Testing

• cmake --build --preset default --target lmdblib_tests -j$(nproc)
• ctest --preset default -R 'LMDBStoreTest\.(duplicate_cursor_helpers_throw_on_non_duplicate_database|filtered_get_value_or_previous_returns_false_on_empty_database|get_stats_throws_for_uncommitted_database|write_transaction_rejects_second_commit)$' --output-on-failure
• ctest --preset default -R 'LMDB(Store|Environment)Test\.' --output-on-failure
• cmake --build --preset default --target crypto_merkle_tree_tests -j$(nproc)
• ctest --preset default -R 'LMDBTreeStoreTest\.' --output-on-failure

Issues

• Closes AztecProtocol/barretenberg-claude#2649 — lmdblib error handling and transaction guard failures.
• Refs AztecProtocol/barretenberg-claude#2544 — LMDB M1/M2 failure-handling portion only.

---

Created by claudebox · group: bb-audit-remediation

[AztecBot]

Flakey Tests

🤖 says: This CI run detected 1 tests that failed, but were tolerated due to a .test_patterns.yml entry.

\033FLAKED\033 (8;;http://ci.aztec-labs.com/a88a67c22c5b93bf�a88a67c22c5b93bf8;;�):  yarn-project/end-to-end/scripts/run_test.sh simple src/multi-node/slashing/inactivity_slash_with_consecutive_epochs.test.ts (314s) (code: 0) group:e2e-p2p-epoch-flakes