draft: add ARKit52 NPC voice lip sync lane

[JOY (JOY)]

Summary

• Adds ARKit52/provider lip-sync plumbing for scoped NPC voice sessions.
• Adds local ARKit52 smoke/helper scripts, a sidecar launcher, and blendshape reporting utilities.
• Adds a synthetic Unity editor ARKit52 driver smoke hook that feeds provider-style frames into PrototypeFacialAnimationDriver and checks actual SkinnedMeshRenderer blendshape weights.
• Adds visual prefab catalog safety checks and documents the current missing paid-source prefab state.
• Keeps the lane presentation-only: voice and lip-sync payloads do not mutate authoritative gameplay state.
• Shows focused NPC text before voice buffering, and stops stale NPC voice presentation when the player sends the next line so chat stays responsive.
• Extends the ARKit52 smoke helper to verify secondspawn_voice_audio_chunk_get, so chunked Gemini audio retrieval is tested instead of only voice-session metadata.
• Adds non-mutating lip-sync provider readiness and Unity lip-sync contract checkers so agents can verify local provider/testability state without touching production services.
• Adds Nakama runtime coverage for Gemini voice gender pool selection and stable actor voice assignment.
• Merged current origin/dev into this draft branch after resolving the voice lane doc conflict.

Verification

• dotnet build Unity\SecondSpawn.AI.csproj --nologo: passed, 0 warnings.
• dotnet build Unity\SecondSpawn.UI.csproj --nologo: passed, 0 warnings.
• dotnet build Unity\Assembly-CSharp-Editor.csproj --nologo: passed. Warnings are existing Photon/Fusion package warnings, not new SECOND SPAWN code.
• npm.cmd run build: passed.
• npm.cmd test: passed after build, including Gemini voice pool selection coverage.
• git diff --check: passed.
• python -m py_compile tools\lipsync\check_unity_lipsync_contract.py tools\lipsync\check_lipsync_provider_readiness.py tools\lipsync\run_arkit52_smoke.py tools\lipsync\wav2arkit_http_server.py: passed.
• python tools\lipsync\check_unity_lipsync_contract.py: passed 13/13 checks for ARKit52 sidecar channel shape, Nakama chunk smoke assertions, Unity DTO shape, presenter lip-sync forwarding, facial driver blendshape mapping, synthetic editor driver smoke hook, focused/free-mode voice gating, and portrait ARKit status reporting.
• python tools\lipsync\run_arkit52_smoke.py: passed, 99 frames, 52 channels, 72.11 ms wall in the isolated worktree.
• python tools\lipsync\run_arkit52_smoke.py --include-nakama --actor-id npc-scrap-warden-0441 --text "Xin chao. Day la spike test ARKit nam muoi hai voi giong nhan vat nam.": passed, provider gemini_tts, transport nakama_audio_chunks, voice Algenib, male hint, 158 ARKit52 frames, 52 channels, 104 ms lip-sync latency, 6580.26 ms total RPC wall time, 3 audio chunks, 335360 base64 chars fetched.
• Attempted Unity.exe -batchmode -nographics -quit -projectPath .claude/worktrees/arkit52-lipsync-root/Unity -executeMethod SecondSpawn.EditorTools.SecondSpawnFacialBlendshapeReportUtility.RunArkit52DriverSmoke: Unity did not reach the smoke method because Package Manager failed startup with The "path" argument must be of type string. Received undefined. This is recorded as a worktree batch runner blocker, not a failed driver smoke.
• Earlier male/female voice smoke: npc-scrap-warden-0441 returned stable voice Algenib twice with male hint; npc-clinic-operator-0320 returned Sulafat with female hint. Both returned Gemini audio chunks plus ARKit52 frames.
• python tools\lipsync\check_lipsync_provider_readiness.py --scratch-root D:\Projects\Second-Spawn\.tmp-import\lipsync-spike: passed. Only wav2arkit_cpu is testable now; SALSA/uLipSync/Convai/FaceSync are not installed in Unity, NeuroSync is blocked by broken torchaudio, and NVIDIA A2F SDK is blocked by CUDA 13.3 plus missing TENSORRT_ROOT_DIR.
• python tools\unity\check_visual_prefab_catalog.py: passed and reported 50 entries, 29 missing generated prefabs, 50 missing source assets, 3 unresolved generated prefab source GUIDs.
• Local review fallback: no provider keys in Unity, Unity only requests presentation tiers through Nakama/Fusion boundary, branch stays presentation-only.

Still Missing Before Ready

• Unity MCP console and Play Mode smoke could not run. Coplay-style tools returned telemetry but no_unity_session; official Unity MCP tools were visible, and the Editor log showed 20 official MCP tools discovered, but Unity_ManageEditor and Unity_ReadConsole timed out after 120 seconds from this agent session.
• Current root Editor is still on dev; this branch is not merged into dev, so the feature is not yet applied to the root game workspace.
• Final visual proof still requires one focused Ida Faber NPC in Play Mode with audible Gemini voice, honest portrait status, and visible ARKit mapped mouth motion.

Related: #139, #288