fix: resolve empty zip issues by preventing WAV header corruption and handling edge cases

fix: empty zip when using effects without vocal isolation

feat: add advanced pitcher and analyzer tools

feat: integrate 4-stem separation, de-reverb, and ui enhancements

fix: rename dataset folder to final_dataset as requested

feat: natively generate XTTS v2 compliant LJSpeech datasets (Mono 24kHz, 3-column CSV, train/val splits)

feat: complete rewrite of audio processing to chunk and stream massive files to prevent OOM on HF Space

feat: auto-detect CUDA GPU for whisper model to drastically improve speed on GPU hardware

fix: explicit garbage collection and model deletion to prevent OOM server restarts

fix: resolve UnboundLocalError when isolate_vocals is unchecked

chore: move backend files into dedicated backend directory and clean up root