The generate_music() function now supports two backends:

1. audiocraft (preferred):
   - Original MusicGen implementation
   - Works on Python 3.12 and lower
   - Falls back to transformers if not available

2. transformers (fallback):
   - Uses HuggingFace transformers library
   - Works on Python 3.13+
   - No spacy/blis dependency issues

The function automatically:
- Tries audiocraft first (if available)
- Falls back to transformers if audiocraft fails or is not installed
- Provides clear error messages if neither backend is available

This allows MusicGen music generation to work on Python 3.13 without
the problematic audiocraft → spacy → thinc → blis dependency chain.