fix(media): stop permanent voice loss after bad-network blip (protocol v2)

A bad UDP packet on a flaky link could permanently wedge the voice path,
unrecoverable even across app restarts. Three defects:

1. Anti-replay window was advanced from the UNAUTHENTICATED header seq
   before the AEAD tag was checked, and not rolled back on failure. One
   corrupted/forged frame shoved recv_highest_ far ahead, after which every
   legitimate frame was rejected as "too old" forever. Reorder to
   replay-check -> authenticate -> update (RFC 3711 3.3); the window now
   moves only after a successful tag check.

2. The wire seq was only the low 16 bits of the nonce counter (zero-extended
   on receive). After 65,536 frames the nonce desynced and all frames failed
   auth. Widen the voice frame seq u16 -> u64 (header 14 -> 20 bytes). The
   core owns all UDP framing, so Swift/C# clients need only a rebuild. This
   is a versioned wire change: VOICECAT_PROTOCOL_VERSION 1 -> 2, handshake
   rejects on mismatch.

3. Server leaked per-session UDP state on disconnect; unregister_session now
   frees udp_endpoints_/udp_tokens_/ssrc_to_session_.

Also add rate-limited dropped-frame logging to MediaRelay so a wedged media
path is observable. New regression tests in test_media_aead.cpp cover the
poison (fails on old code) and the 16-bit wrap. ctest --preset dev
-E external_pcm: 22/22 pass (external_pcm aborts on a pre-existing CoreAudio
shutdown race, unrelated).

PROGRESS.md

@@ -10,6 +10,35 @@ up instantly. Newest status at the top.
 
 ## ▶ Where we left off / next action
 
+- **Done (2026-06-21):** **Fix permanent voice-loss bug + harden the UDP media path (protocol v2).**
+  Field report: two iOS users lost all audio mid-call after a bad-network blip and could not
+  recover even by restarting the apps. Root causes found in the UDP media path:
+  1. **Anti-replay window poisoned by unauthenticated packets (the trigger).**
+     `SodiumMediaCrypto::open()` advanced `recv_highest_` from the plaintext header `seq`
+     *before* verifying the AEAD tag and never rolled it back on failure. One corrupted/forged
+     frame (a bit-flip on flaky wifi) shoved the high-water mark far ahead, after which every
+     legitimate frame was rejected as "too old" — permanently. Fixed by reordering to
+     replay-check → authenticate → update (RFC 3711 §3.3): the window is now touched only after
+     a successful tag check. Regression test in `test_media_aead.cpp`
+     (`test_corrupted_seq_does_not_poison_window`) — fails on the old code, passes now.
+  2. **16-bit seq wrap with no rollover counter.** The wire header carried only the low 16 bits
+     of the nonce counter (zero-extended on receive); after 65,536 frames the reconstructed
+     nonce diverged and all frames failed auth. **Wire format widened to a full u64 seq**
+     (`voice_frame.h`: header 14 → 20 bytes, `seq` u16 → u64; `crypto.cpp`, `client.cpp`,
+     `media_relay.cpp` updated; `JitterBuffer::Frame::seq` widened). This is a **versioned wire
+     change → `VOICECAT_PROTOCOL_VERSION` 1 → 2**; the `Hello` handshake rejects on mismatch
+     (`conn_session.cpp`). The voice frame is parsed only in `core/`+`server/`+`tests/`, so the
+     Swift/C# clients need only a rebuild — no parser changes.
+  3. **Server leaked UDP state on disconnect.** `SessionRegistry::unregister_session()` now also
+     frees `udp_endpoints_`/`udp_tokens_`/`ssrc_to_session_` (scan-and-erase by session id).
+  4. **Diagnostics.** `MediaRelay` now emits rate-limited dropped-frame counters
+     (unmapped-endpoint / no-recv-crypto / open-failed) so a wedged media path is observable.
+  - **Verified:** `cmake --build --preset dev` clean; `ctest --preset dev -E external_pcm`
+    **22/22 pass** (incl. `m2_voice` e2e relay + the two new AEAD regressions). `external_pcm`
+    still aborts on the **pre-existing** CoreAudio shutdown mutex race (confirmed identical on a
+    clean baseline checkout under the same harness — unrelated to these changes). Docs updated:
+    `voice.md` §2 (header), `protocol.md` (v2 + negotiation), `security.md` (authenticate-then-advance).
+
 - **Done (2026-06-21):** **Expose all channel codec params + guest nickname in every client.**
   - **DRED everywhere + ABI fix.** `dred` (Opus 1.6 Deep REDundancy) existed in the C ABI
     (`vc_audio_config.dred`) and proto but was absent from *both* client marshaling layers — a