[Audit Sink][PART1] Introduce Audit Sink Trait

[HatemMn]

reminder of the order

ci/audit-jsonl-compat
└─ feat/audit-sink-refactor (4 commits)
└─ feat/audit-sink-generalize (1 commit)
└─ feat/audit-postgres-backend (2 commits)
└─ feat/audit-postgres-cli (3 commits)

---

Overview

This PR is just a refactoring/a split for the 2090-line audit file store into three modules before touching anything real:

file_sink.rs (main file), writer.rs (the writer loop), store.rs (the handle + tests), recovery.rs (recovery).

Also pulled event construction into one place in access, split the fat middlewares/audit.rs into mod.rs + client_ip.rs + extensions.rs, and added the new AuditSink trait in interfaces (unused so far).

No behavior change anywhere in this PR.

You might notice that two audit traits are living separatly: one of them is the legacy file trait, it will be removed in the next PR

[Manuthor]

Suggested change

	pub struct ChainHead {
	#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
	pub struct ChainHead {

Derive PartialEq, Eq, and Default on ChainHead so instances can be directly compared (e.g. head == ChainHead::EMPTY) and defaulted in consumers and unit tests.

[Manuthor]

Suggested change

	//! the same canonical hash (see `cosmian_kms_access::audit::canonical_bytes`).
	//! the same canonical hash (see `cosmian_kms_access::audit::compute_row_hash`).

canonical_bytes is private (pub(crate)) inside cosmian_kms_access::audit::hash and not re-exported. Public callers and documentation should reference compute_row_hash or cosmian_kms_access::audit.

[Manuthor]

Suggested change

	RowCheck::Verified(event) => match event.id.checked_add(1) {
	match check_row(anchor_line) {
	RowCheck::Verified(event) => {
	if !verify_chain_link(&event, previous_event) {
	return Ok(TailOutcome::SealAndRoll {
	reason: SealReason::ChainBroken,
	claimed_last_id: Some(event.id),
	failure_offset: last_start,
	});
	}
	match event.id.checked_add(1) {
	Some(next_id) => Ok(TailOutcome::TruncateContinue {
	keep_len: last_start,
	next_id,
	prev_hash: event.row_hash,
	bytes_discarded: last_end - last_start,
	discard_offset: last_start,
	}),
	None => Ok(TailOutcome::SealAndRoll {
	reason: SealReason::IdOverflow,
	claimed_last_id: Some(event.id),
	failure_offset: last_start,
	}),
	}
	}
	RowCheck::HashMismatch(_) | RowCheck::Unparsable => Ok(TailOutcome::SealAndRoll {

When recovering from a torn trailing write, verify verify_chain_link(&event, previous_event) on the fallback anchor event before accepting it as the truncation boundary. If the fallback anchor row itself has a broken hash link, classify_tail should trigger SealAndRoll rather than resuming over a corrupted chain.