Add Nucleotide Transformer v2 SeqTrainer tutorial for promoter tasks

[Gonza10V]

Motivation

• Make it easy to use the InstaDeep Nucleotide Transformer v2 from within SeqTrainer by providing a package-native backbone spec and convenience accessors.
• Demonstrate realistic promoter tasks (binary classification and regression) using SeqTrainer dataset abstractions together with Hugging Face Trainer to serve as a reproducible example for users.

Description

• Add src/seqtrainer/torch/hf.py which defines NucleotideTransformerV2Spec and provides register_default_torch_backbones and get_nucleotide_transformer_v2_backbone to register/resolve a built-in backbone spec for InstaDeepAI/nucleotide-transformer-v2-500m-multi-species.
• Export the new helpers from seqtrainer.torch by updating src/seqtrainer/torch/__init__.py so the backbone can be resolved via from seqtrainer.torch import get_nucleotide_transformer_v2_backbone.
• Add tutorial notebook notebooks/tutorials/06_ntv2_promoter_tasks.ipynb that loads the example promoter dataset, builds SeqTrainer MaterializedDataset splits, tokenizes with the resolved tokenizer, and shows HF Trainer setup for both classification (median-thresholded) and regression (raw target); training/eval calls are commented to avoid accidental heavy runs.
• Update README.md to include the new notebook in the tutorial list.

Testing

• Ran python -m compileall src/seqtrainer to ensure files compile and the package files are syntactically valid, which succeeded.
• Verified the backbone helper via PYTHONPATH=src python - <<'PY' ... get_nucleotide_transformer_v2_backbone() ... PY, which returned the expected BackboneSpec successfully.
• Confirmed that importing without an editable install fails in a plain interpreter (environment limitation), so the recommended usage is via editable install or PYTHONPATH=src; this import failure is environment-specific and not a code defect.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ba4856a8a2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use evaluation_strategy for Transformers 4.30 compatibility

This notebook passes eval_strategy to TrainingArguments, but this parameter is not accepted in the project's declared minimum transformers versions (e.g., 4.30.x uses evaluation_strategy), so running the tutorial in a compliant environment raises TypeError before training starts. The same incompatible argument appears in both classification and regression argument blocks, which makes the tutorial fail for users installing seqtrainer[torch] with the documented dependency floor.

Useful? React with 👍 / 👎.