Remove Use AI enhancement toggle — always use AI for character extraction since Qwen3-TTS is the default engine

Polish pass: add File download for ZIP support, Clear button, sample load updates stats/chapters, GPU detection cleanup, character descriptions persist in save/load, long-text warnings, first-run notes on Generate tab.

Robust GPU detection: TTSEngine checks torch.cuda.is_available() at model load time, with CPU-safe float32 fallback.

AI-by-default: narrator and characters default to design mode. AI enhancement checkbox defaults ON. AI extraction returns voice descriptions. Fix load project output count.

Major enhancement: file upload, chapter detection, segment previews, multi-format export, speed control, project save/load, quick generate, sample stories

Initial deploy: AudioBook Forge with Qwen3-TTS backend, character voice mapping, and dark Gradio UI