Add DeBERTa training and ONNX export pipeline for required phrases extraction

[Kaushik-Kumar-CEG]

Part of #5137

adds the DeBERTa BIOES/CRF training and export pipeline for required phrase extraction

the training command:

• validates the dataset and BIOES labels before training
• uses a pinned model revision and reproducible runtime settings
• selects the checkpoint with the best validation F1
• reloads and verifies the final model offline
• writes SUCCESS.json only after all artifact checks pass
• exports constrained CRF data and optional ONNX emissions

Final run results:

• validation F1: 0.9008
• test F1: 0.8766
• test precision: 0.8748
• test recall: 0.8784
• test exact match: 0.8197
• invalid BIOES paths: 0

the uploaded model was downloaded and validated again successfully.

The package version is in aboutcode-org/scancode-required-phrases#2.

disclosure : used Claude to help review and clean up few bugs in script

[Shriprasad-P]

Review

PR: Add DeBERTa training and ONNX export pipeline for required phrases extraction

Touched: etc/requirements-ml.txt, etc/scripts/dataset_pipeline/export_onnx.py, etc/scripts/dataset_pipeline/phrase_model.py, etc/scripts/dataset_pipeline/test_export_onnx.py, etc/scripts/dataset_pipeline/test_phrase_model.py, etc/scripts/dataset_pipeline/test_train_model.py

• Addresses a concrete correctness/reliability issue based on the title and diff.
• CI/tooling change — confirm the pipeline still passes on this branch.
• Diff is fairly large (+5133/-0); a short summary of risk areas from the author would help reviewers.

Commenting as a drive-by reviewer after reading the diff. Happy to look again if maintainers want a deeper pass on a specific file.