docs: add build/manual-testing guide, fix stale M0 stub claims in headers

- docs/building.md: explains what each CMake preset (dev, m1-dev, m2-dev,
  server-release) is actually for, and how to build voicecat-server + vccli
  for manual testing. Linked from CLAUDE.md's doc index.
- core/include/voicecat.h, core/src/voicecat.cpp, core/src/protocol/protocol.h,
  server/src/main.cpp: doc-header comments still claimed M0-skeleton/stub
  behavior (VC_ERR_NOT_IMPLEMENTED everywhere, "prints what it would do",
  protobuf codegen "commented") that M1-M3 made real. Updated to describe
  current behavior, with the dev-preset stub fallback noted explicitly where
  it still applies.

core/include/voicecat.h

@@ -8,8 +8,13 @@
  * Design: docs/architecture.md §4. Everything here is async + event-driven — calls return
  * immediately and results/state changes arrive via the vc_callbacks.on_event callback.
  *
- * STATUS: M0 skeleton. Implementations live in core/src and currently return
- * VC_ERR_NOT_IMPLEMENTED. The shapes below are the contract to build against.
+ * STATUS: real, behind VOICECAT_HAS_NET (the `m1-dev`/`server-release` presets — vcpkg deps
+ * on; see docs/building.md). As of M3, control plane, voice, multi-stream, device
+ * enumeration, VAD/PTT, and stereo playback all work for real via core/src/core/client.cpp.
+ * The no-deps `dev` preset still links a stub vc_client that returns VC_ERR_NOT_IMPLEMENTED
+ * for everything below `connect`, purely to keep that skeleton build green. webrtc AEC/NS/AGC
+ * remains an inert passthrough regardless of preset (no Windows/MSVC port upstream —
+ * docs/voice.md §8/§11, PROGRESS.md).
  */
 #ifndef VOICECAT_H
 #define VOICECAT_H

core/src/protocol/protocol.h

@@ -6,7 +6,10 @@
  * request_id ↔ response, and dispatches to handlers. Media frames do NOT come through here
  * (they use the fixed binary header in voice.md §2).
  *
- * STATUS: M0 stub — protobuf codegen is wired in CMake (commented) and turned on in M1.
+ * STATUS: real. Protobuf codegen is on (core/CMakeLists.txt) for VOICECAT_HAS_NET builds
+ * (`m1-dev`/`server-release`); FrameCodec below is fully implemented and used by both the
+ * client (net/transport.h) and the server (conn_session.cpp). See protocol/envelope.h for the
+ * Envelope-level encode/decode that sits on top of this.
  */
 #ifndef VOICECAT_PROTOCOL_PROTOCOL_H
 #define VOICECAT_PROTOCOL_PROTOCOL_H

core/src/voicecat.cpp

@@ -1,9 +1,11 @@
 /*
- * voicecat.cpp — C ABI implementation (M0 skeleton).
+ * voicecat.cpp — C ABI implementation.
  *
- * Lifecycle (create/destroy) and trivial accessors are real. Everything that needs a
- * subsystem (net/crypto/codec/protocol/session/audio) returns VC_ERR_NOT_IMPLEMENTED for
- * now and is the work of M1+ (see AGENTS.md / docs/roadmap.md).
+ * Lifecycle (create/destroy) and trivial accessors are always real. Everything else below
+ * just delegates to vc_client (core/src/core/client.cpp): under VOICECAT_HAS_NET
+ * (`m1-dev`/`server-release` — see docs/building.md) that's the real M1–M3 implementation;
+ * under the no-deps `dev` preset, client.cpp's `#else` branch returns VC_ERR_NOT_IMPLEMENTED
+ * for all of it, to keep that skeleton build green.
  */
 #include "voicecat.h"
 
@@ -48,7 +50,7 @@ vc_client* vc_client_create(const vc_config* cfg, vc_callbacks cb) {
 
 void vc_client_destroy(vc_client* c) { delete c; }
 
-/* ── Everything below delegates to the (stub) client. ─────────────────────── */
+/* ── Everything below delegates to vc_client (real or stub, per preset above). ───────── */
 
 vc_result vc_connect(vc_client* c, const char* host, uint16_t port) {
     if (c == nullptr || host == nullptr) return VC_ERR_INVALID_ARG;