feat(voice): add Qwen3-TTS voice design, consented cloning, and fine-tuning (#5381)

[atomantic]

Closes #5381

Summary

Adds Qwen3-TTS local voice backend integration across the voice engine contract, voice profiles, and Universe Voice Lab:

• Isolated Python Runtime & Probing: Adds health probing, hardware detection (CUDA / Apple Silicon MPS / CPU), model management, and explicit on-demand downloads (server/services/voice/qwen3TtsRuntime.js, scripts/qwen3_tts_runner.py). No boot-time cold inference or downloads.
• Voice Design: Instruction- and seed-controlled candidate voice generation via Qwen3-TTS 1.7B Voice Design model.
• Consented Single-Speaker Cloning: Rapid cloning with machine-local reference audio, SHA256 integrity hashing, exact transcript conditioning, and required performer consent confirmation. Reference audio remains strictly machine-local.
• Voice Fine-Tuning: Optional, cancellable, checkpointed fine-tuning lifecycle. Emits intermediate checkpoints with audition samples and supports explicit promotion of any selected checkpoint.
• Interactive Routing Qualification: Evaluates first-audio latency benchmark on the local host before enabling the interactive route (studio remains separately gated).
• Voice Lab UI: Adds tabbed Voice Lab interface in CharacterDetailEditor for preset promotion, voice design, consented cloning, fine-tuning management, and benchmark audition.

Key Changes

• scripts/qwen3_tts_runner.py: Standalone CLI runner for runtime probing, voice design inference, consented cloning, standard synthesis, and step-based fine-tuning.
• server/services/voice/qwen3TtsRuntime.js: Probing, GPU/hardware capability inspection, model catalog, and user-initiated downloads.
• server/services/voice/tts-qwen3.js: Qwen3-TTS synthesis adapter handling prompt conditioning, reference alignment, and rate controls.
• server/services/voice/fineTuning.js: Machine-local dataset validation, background training job runner, progress tracking, and checkpoint promotion.
• server/services/voice/profiles.js: Support for designed, cloned, and fine-tuned profile kinds with explicit candidate drafting and approved promotion.
• server/services/voice/profileBenchmarks.js: Latency benchmark qualification for interactive speech.
• server/routes/voice.js: REST endpoints for design, clone, promotion, fine-tuning lifecycle, and Qwen3 status/downloads.
• client/src/components/universe/CharacterDetailEditor.jsx: Full Voice Lab UI supporting design previews, consent verification, fine-tuning controls, and candidate audition.
• client/src/services/apiVoice.js: Extended API wrappers for all new voice endpoints.

Testing

• server/services/voice/qwen3TtsRuntime.test.js: Validates model registry, health probing, and download requests.
• server/services/voice/tts-qwen3.test.js: Validates voice design synthesis and effective controls.
• server/services/voice/profiles.test.js: Validates draft candidate creation, mandatory consent enforcement, and checkpoint promotion.
• server/services/voice/fineTuning.test.js: Validates dataset verification, training progress, checkpoint generation, and job cancellation.
• server/routes/voice.test.js: Validates all new REST endpoints and upload cap enforcement.
• client/src/components/universe/CharacterDetailEditor.test.jsx: Validates Voice Lab tabs, consent gating, and interactive qualification.
• Full client and server test suites run cleanly.