fix understand_audio: clone tensors for inference mode, working on GPU (52s)

add understand_audio (LM reverse), demucs-infer fix, commit refs, dtype fixes

major update: PyTorch inference, Gradio 6, session isolation, /understand captioning

SDPA first on Blackwell, FA2 only for Ampere/Hopper, txt caption support

add GPU/CUDA auto-detect, mixed precision, flash_attn, txt caption parser

save only on cancel/finish, max epochs 1000, default 3

add mid/sas analysis modes (Demucs + ensemble), auto-select by dataset size

add auto-captioning (BPM/key/signature via librosa), add librosa+mutagen deps

add XL variant mappings to _VARIANT_DIR

fix: adapter saved to clean dir, LM dropdown no 'Default', on-demand download

Side-Step training engine, tested locally on CPU