feat: Add late interaction model training support for retrieval

[rnyak]

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Updated train_bi_encoder.py to support local ColBERT-style pooling by adding colbert_scores_and_labels(), which computes MaxSim scores with query and passage attention-mask handling. The train and validation paths now route ColBERT models through this scoring function instead of standard pooled embedding contrastive scoring.

Distributed in-batch negatives remain unsupported for ColBERT for now and still raise explicitly.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[rnyak]

/claude review