fix: ignore an ACK whose CSeq does not match the server INVITE

[tgeorge06]

Fixes #154.

Problem

A server INVITE transaction in Completed or Confirmed moves to Confirmed on any ACK (transaction.rs, the Completed | Confirmed if req.method == Method::Ack arm of on_received_request); it never checks the ACK's CSeq.

An ACK for a 2xx has its own branch, so EndpointInner::on_received_message routes it per dialog through waiting_ack. A delayed ACK of an earlier INVITE or re-INVITE on the same dialog therefore reaches the transaction of the current re-INVITE. That transaction then stops Timer G (a lost 2xx is never retransmitted), removes its waiting_ack entry (the real ACK loses its route and is dropped), and hands the stale ACK to the TU as this re-INVITE's ACK.

Spec

• RFC 3261 §17.1.1.3: the ACK for a non-2xx has the same CSeq number as the INVITE.
• RFC 3261 §13.2.2.4: the CSeq number of the ACK for a 2xx equals the INVITE's.
• RFC 3261 §13.3.1.4: the 2xx is retransmitted until the ACK for it arrives.

Fix

In that arm, compare the ACK's CSeq number with self.original's. On a mismatch, log at debug level and return None before touching any state: no transition, Timer G and the waiting_ack route stay as they are, nothing goes to the TU. 16 lines in src/transaction/transaction.rs, plus tests. No public API change.

Contract / coverage

• ACKs that follow the RFC are unaffected: the hop-by-hop ACK for a non-2xx (same branch) and the ACK for a 2xx both carry the INVITE's CSeq number, so both still confirm. A retransmitted matching ACK behaves as before.
• A request with a missing or unparseable CSeq never reaches this arm (TransactionKey::from_request rejects it).
• Transport-agnostic: the check runs before any timer handling, so it covers UDP (Timer G keeps running) and reliable transports alike. CANCEL is handled earlier in the same function and is unaffected; the multi-Via 2xx handling from 3ea35cd is on the client side and unrelated.
• Compatible with fix: end the session when a UAS 2xx to an INVITE is never ACKed #149 (2xx retransmission and ACK wait on every transport): the two touch different lines, merge without conflict, and the combined tree passes cargo test --lib.
• Orthogonal to feat(transaction): RFC 6026 Accepted state + Timer L/M for INVITE 200 OK retransmission absorption #128 (RFC 6026 Accepted state): feat(transaction): RFC 6026 Accepted state + Timer L/M for INVITE 200 OK retransmission absorption #128 keeps the same unchecked ACK arm and its Accepted arm also forwards any ACK to the TU. If feat(transaction): RFC 6026 Accepted state + Timer L/M for INVITE 200 OK retransmission absorption #128 lands, the same CSeq-number guard belongs in the Accepted arm; happy to rebase either way.

Left out to keep this minimal (happy to follow up):

• The CSeq method is not checked: an ACK carrying CSeq: 2 INVITE is still accepted, as today.
• The finished_transactions ACK path in on_received_message (transaction already dropped by the TU) removes the waiting_ack entry without a CSeq check. The ACK is absorbed silently there either way and nothing is retransmitted any more, so there is no observable effect.
• waiting_ack holds one entry per dialog, so a delayed ACK of re-INVITE N is now ignored rather than routed back to a still-live transaction N (the N+1 entry already overwrote N before this change).

Tests

New src/transaction/tests/test_server_invite_ack.rs (real sockets and the endpoint serve loop, same pattern as test_server_invite_drop.rs):

• test_server_invite_ignores_ack_with_other_cseq (UDP): re-INVITE CSeq: 2 answered 200; a delayed CSeq: 1 ACK is not delivered, the state stays Completed, Timer G keeps running, the waiting_ack route survives and the 200 is retransmitted; the CSeq: 2 ACK then confirms, stops Timer G and clears waiting_ack.
• test_server_invite_ignores_ack_with_other_cseq_over_tcp: same dialog over TCP; the CSeq: 1 and CSeq: 2 ACKs are written back to back on the ordered stream, and the first ACK the transaction delivers is the CSeq: 2 one, which confirms it and clears waiting_ack.

Checks (on main @ 3ea35cd)

• Red: with the src/ change reverted and the tests kept, cargo test --lib test_server_invite_ack → 2 failed (an ACK with CSeq 1 must not be delivered by the CSeq 2 INVITE transaction, got Some("ACK … CSeq: 1 ACK …") over UDP; left: 1, right: 2 over TCP).
• Green: cargo test → 334 lib + 65 doc passed, 0 failed (main: 332 lib). rustfmt --check clean on the changed files.