Add documentation to MusicBertNoteEncoder class

[Copilot]

Addresses review feedback requesting documentation for the MusicBertNoteEncoder class to explain its architecture and two-stage encoding process.

Changes

• Class-level docstring: Documents the two-stage architecture (MusicBERT token embedding → token-to-note pooling) and parameter usage including LoRA adapter configuration
• Forward method docstring: Specifies input/output tensor shapes and the token-to-note alignment mechanism

The documentation clarifies how token-level MusicBERT embeddings are aggregated into note-level representations using weighted pooling based on alignment edges.

class MusicBertNoteEncoder(nn.Module):
    """Note-level encoder using MusicBERT with token-to-note pooling.
    
    This encoder implements a two-stage process for generating note-level representations
    from tokenized musical sequences:
    
    1. Token Embedding: A pretrained MusicBERT backbone (transformers AutoModel) encodes
       the input token sequence into contextualized token embeddings.
    2. Token-to-Note Pooling: A TokenToNotePooler aggregates token embeddings into 
       note-level representations using weighted pooling based on token-to-note alignments.
    ...
    """

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.