chore: configure HuggingFace Spaces deployment

centralized datasets inside data, removed hardcoded paths and updated paths for different OS. Updated rebuild_bilingual_corpus.py and train_xlm_bilingual_bilingual.py.

feat: add bilingual support with XLM-RoBERTa and classical bilingual models

add colab notebook, distilbert, xai, inference module

checkpoint 2: classical models, streamlit app, visualisations