feat(tts): Basic TTS model loading and inference

[m96-chan]

Overview

Implement basic TTS (Text-to-Speech) functionality in PyGPUkit.

Goals

• Select TTS model (e.g., VITS, Piper, Kokoro, etc.)
• Implement SafeTensors model loading
• Basic text-to-audio inference
• Output to PCM/WAV

Technical Considerations

• Model format: SafeTensors preferred
• Output: GPU-side audio buffer (GPUArray)
• Dtype: float32 or int16 PCM

Acceptance Criteria

• Model loads without error
• Text input → Audio output works
• Basic quality validation

Related

• Part 1 of TTS pipeline implementation
• Prerequisite for #(LLM→TTS) and #(Full Voice Pipeline)