feat(transaction): RFC 6026 Accepted state + Timer L/M for INVITE 200 OK retransmission absorption

[andrico21]

Implements RFC 6026 §7.1–§7.2 to fix the long-standing issue where rsipstack's INVITE state machine terminates immediately on 2xx and cannot absorb retransmitted 200 OKs from the UAS, breaking glare resolution and forking scenarios.

Closes #127. Per maintainer @shenjinti's confirmation on the issue thread: "you're completely right ... A PR would be very welcome."

Branch is rebased onto current upstream/main (8928174).

Standards

• RFC 6026 §7.1 (server INVITE 2xx → Accepted; Timer L = 64·T1; absorbs ACK retransmits + late forked 2xx)
• RFC 6026 §7.2 (client INVITE 2xx → Accepted; Timer M = 64·T1; absorbs retransmitted 2xx and forwards each to TU; client does NOT auto-ACK in Accepted — caller TU/dialog layer is responsible per strict RFC)
• RFC 3261 §17.1.1 / §17.2.1 (legacy Completed semantics preserved for 3xx-6xx)

Changes

Area	File
Accepted state variant	src/transaction/mod.rs
TimerL + TimerM variants	src/transaction/mod.rs
can_transition matrix	src/transaction/mod.rs
timer_l + timer_m fields	src/transaction/transaction.rs
Accepted-entry handler	src/transaction/transaction.rs
on_timer Accepted dispatch	src/transaction/transaction.rs
Server-side 2xx routing	src/transaction/transaction.rs::respond
Client-side 2xx routing	src/transaction/transaction.rs::on_received_response
9 conformance tests	src/transaction/tests/test_transaction_states.rs

Both TransactionState and TransactionTimer are now #[non_exhaustive] to allow future RFC additions without breakage. This is itself a breaking change — recommend targeting 0.6.0.

Backward compatibility

The pre-existing convenience that auto-ACKs client INVITE 2xx from on_received_response is retained for 0.5.x consumers (technically a deviation from RFC 3261 §17.1.1.3, which mandates the TU drive the 2xx ACK). Documented honestly in lib.rs::Standards Compliance. Removal is a candidate for a follow-up PR that migrates 2xx-ACK responsibility to the dialog layer.

Validation

• cargo test --lib — 247 passed, 0 failed
• cargo test --doc — 65 passed, 0 failed
• cargo fmt --check — clean

Pre-existing baseline: 66 clippy errors on main (not introduced by this PR; touched files contribute zero new clippy regressions).

Pre-submit review

Two read-only architecture review passes were run before submission. Each finding was either fixed in an atomic follow-up commit or explicitly deferred with rationale.

First pass — 3 MUST-FIX + 4 RECOMMEND + 3 NIT:

• F1: pass every 2xx in Accepted to TU (skip dup-suppression per §7.2)
• F2: lib.rs auto-ACK retention documented honestly
• F3: #[non_exhaustive] on both public enums
• R2: log auto-ACK failures (was silent .ok())
• R3: cancel-safety rustdoc on respond / send_ack / receive
• R4: typed (transaction_type, timer) dispatch in Accepted entry handler
• N1: explicit stray-timer list with warn! on mismatched pairing
• N2: dropped internal commit-hash references from comments

Second pass — 4 SHOULD-FIX (run after rebase onto current main, evaluated against an internal Rust style guide):

• D1: respond cancel-safety rustdoc had inverted mutation order — fixed to describe actual store last_response → await send → transition order
• D2: send_ack rustdoc claimed a nonexistent self.connection fallback — rewritten to describe actual two-source resolution and silent no-send fallback when neither input arg nor 2xx-Contact lookup yields a connection
• D3: inline comment in client Accepted-state entry contradicted the retained auto-ACK gate — rewritten to match the lib.rs disclaimer
• T1: dead wildcard arm in send_ack post-send dispatch replaced with if/else + debug_assert_eq! invariant (catches future #[non_exhaustive] additions)

Follow-up out of scope for this PR

• R1: client-side fast Timer L/M tests via T1 override knob (~100 LOC of test-endpoint instrumentation; happy to ship as a follow-up if requested)
• 2xx-ACK migration to dialog layer (removes the §17.1.1.3 deviation)

Commit list (atomic, 20 commits)

1. feat(transaction): add Accepted state for RFC 6026 §7.1/§7.2
2. feat(transaction): add TimerL + TimerM variants for RFC 6026
3. feat(transaction): allow RFC 6026 edges in can_transition matrix
4. feat(transaction): add timer_l + timer_m fields on Transaction
5. feat(transaction): Accepted-state entry handler per RFC 6026 §7.1/§7.2
6. feat(transaction): on_timer Accepted dispatch + Completed 2xx guard
7. feat(transaction): server-side RFC 6026 2xx routing + ACK handling
8. feat(transaction): client-side RFC 6026 2xx routing + send_ack split
9. test(transaction): RFC 6026 conformance tests for Accepted state
10. docs(lib): expand RFC 6026 standards-compliance claim
11. style: cargo fmt --all cleanup
12. fix(transaction): pass every 2xx in Accepted to TU per RFC 6026 §7.2 [F1]
13. docs(lib): clarify RFC 6026 client-side auto-ACK retention [F2]
14. feat(transaction): mark TransactionState + TransactionTimer #[non_exhaustive] [F3]
15. refactor(transaction): tighten Accepted timer dispatch [R4 + N1 + N2]
16. fix(transaction): log auto-ACK failures instead of silent .ok() [R2]
17. docs(transaction): cancel-safety rustdoc on async public APIs [R3]
18. docs(transaction): correct rustdoc and comment accuracy on touched APIs [D1 + D2 + D3]
19. refactor(transaction): drop dead wildcard arm in send_ack post-send dispatch [T1]
20. style: cargo fmt cleanup on rebased F1 hunk

[shenjinti]

20 commits is too granular for a single feature. Please squash into ~3-5 logical commits (e.g. feat, docs, test). The style: cargo fmt cleanup commits and (N/N) suffix noise can be folded into their parent commits.

The PR body mentions auto-ACK retention as a known RFC deviation. Consider adding an EndpointBuilder option (e.g. .auto_ack_invite_2xx(bool), default true) so users can opt into strict RFC 3261 §17.1.1.3 behavior without a breaking change. This can be a follow-up, not a blocker for this PR.

The on_timer → Accepted → Terminated path has no integration test (deferred as R1). Consider adding a t1_override test knob so this can be tested without waiting 32s.

[andrico21]

Squashed to 4 commits per your spec. Split the breaking #[non_exhaustive] change out as its own commit for bisect precision; everything else folded as you asked. Force-pushed:

• feat(transaction): RFC 6026 Accepted state + Timer L/M for INVITE 2xx absorption
• feat(transaction)!: mark TransactionState + TransactionTimer #[non_exhaustive]
• test(transaction): RFC 6026 conformance tests for Accepted state
• docs(transaction): RFC 6026 standards-compliance + cancel-safety + rustdoc accuracy

(N/N) suffixes + style fmt commits folded into parents. Final state matches the previous PR HEAD: 247 lib tests pass, fmt clean.

On .auto_ack_invite_2xx(bool) builder option: agreed - will land as a follow-up PR after this merges.

On the t1_override test knob: can do as follow-up too. Want it folded in before merge or its own PR?