test(audit): expand buzz-audit unit test coverage (12 → 32)

[jewoos2921]

Summary

The audit crate's core chain logic had thin unit test coverage — most existing service.rs tests require Postgres and are #[ignore]'d, leaving the hash-chain integrity logic under-tested in just test-unit.

This PR adds 20 new DB-free unit tests, bringing runnable coverage from 12 → 32 tests.

What's covered

hash.rs (+12 tests)

canonical_json depth:

• Nested object key sorting (recursion into sub-objects, not just top level)
• Array order preservation (arrays are sequences, not sets)
• Empty values ({}, [], null, scalars)
• Special-character escaping with round-trip verification

compute_hash edge cases:

• object_id presence tag (None vs Some("") — the same invariant actor_pubkey has)
• detail None-vs-empty-object distinction
• detail key-order invariance (canonical JSON working as designed)
• detail value sensitivity
• created_at field sensitivity (1µs shift changes hash)
• Sub-microsecond jitter invariance (500ns truncates away — the whole point of to_storage_precision)
• GENESIS_HASH is all-zero sentinel
• prev_hash = None hashes identically to prev_hash = Some(GENESIS_HASH)

action.rs (+4 tests)

• serde JSON roundtrip for all variants
• serde snake_case agrees with as_str() (catches variant drift)
• as_str() values are unique (no ambiguous DB strings)
• as_str()/FromStr are exact inverses

service.rs (+3 tests)

• log_timestamp is idempotent under re-truncation (write/read preimage safety)
• Advisory lock key format is namespaced + community-scoped
• Different communities produce different lock keys (per-community serialization invariant)

Verification

• cargo test -p buzz-audit --lib — 32 passed, 6 ignored (Postgres), 0 failed
• cargo clippy -p buzz-audit -- -D warnings — clean
• cargo fmt --check -p buzz-audit — clean