feat(M2): UDP voice/media plane -- SFU relay, Opus, AEAD, jitter buffer

Adds the full voice pipeline: 14-byte binary frame header, ChaCha20-Poly1305
AEAD keyed from the TLS exporter, libopus encode/decode with FEC/PLC/DTX,
an adaptive per-ssrc jitter buffer, a miniaudio capture/playback engine, an
APM passthrough stub, and the UdpBinding/StreamAnnounce signaling chain
wired through ConnSession/SessionRegistry into a new server-side SFU
(MediaRelay) that decrypts and re-encrypts frames per channel member.

Exit criterion verified: test_m2_voice — two headless clients relay 50
encrypted Opus frames through the server; ctest --preset m1-dev is 9/9
green. Also corrects protocol.md's UdpBinding diagram, which described the
UDP-side binding packet as AEAD-sealed when it is in fact a plaintext
bootstrap frame (separate from the TCP/TLS UdpBinding ack).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

server/src/media_relay.h

@@ -0,0 +1,65 @@
+/*
+ * server/media_relay.h — UDP SFU relay for M2 voice.
+ *
+ * Design: docs/architecture.md §5, docs/voice.md §2.
+ * Receives encrypted UDP voice frames from clients, decrypts+authenticates them,
+ * re-encrypts for each channel member, and forwards.
+ *
+ * Flow:
+ *   1. Client sends kFrameUdpBinding UDP packet (plaintext) → MediaRelay looks up
+ *      the 16-byte token in SessionRegistry, associates the sender endpoint with
+ *      the ConnSession, and calls session->set_udp_endpoint().
+ *   2. Client sends kFrameVoice UDP packets → MediaRelay decrypts via recv_crypto(),
+ *      finds channel members via find_channel_sessions(), re-encrypts via send_crypto(),
+ *      and sends to each member's UDP endpoint.
+ */
+#ifndef VOICECAT_SERVER_MEDIA_RELAY_H
+#define VOICECAT_SERVER_MEDIA_RELAY_H
+
+#ifdef VOICECAT_HAS_NET
+
+#include <memory>
+
+#define ASIO_STANDALONE 1
+#include <asio.hpp>
+
+#include "net/transport.h"
+
+namespace voicecat::server {
+
+class SessionRegistry;
+
+class MediaRelay {
+   public:
+    MediaRelay(asio::io_context& io, std::shared_ptr<SessionRegistry> registry);
+    ~MediaRelay();
+
+    // Bind the UDP socket. port=0 lets the OS pick. Must be called before start().
+    bool bind(uint16_t port = 0);
+
+    // Begin async receive loop. Call once after bind().
+    void start();
+
+    // Stop receiving and close the socket.
+    void stop();
+
+    // Actual bound port (after bind()).
+    uint16_t media_port() const;
+
+   private:
+    void on_udp_frame(const uint8_t* data, size_t len, asio::ip::udp::endpoint sender);
+
+    asio::io_context&                io_;
+    std::shared_ptr<SessionRegistry> registry_;
+    voicecat::net::UdpMediaChannel   udp_;
+
+    // Scratch buffer for re-encrypted payloads (size = max_frame + 16 MAC)
+    static constexpr size_t kMaxPayload = 1500;
+    std::vector<uint8_t>    seal_buf_  = std::vector<uint8_t>(kMaxPayload + 16, uint8_t{0});
+    std::vector<uint8_t>    plain_buf_ = std::vector<uint8_t>(kMaxPayload, uint8_t{0});
+};
+
+}  // namespace voicecat::server
+
+#endif  // VOICECAT_HAS_NET
+#endif  // VOICECAT_SERVER_MEDIA_RELAY_H