feat: add training and configuration scripts for new run

[Ki-Seki]

No description provided.

[Copilot]

Pull Request Overview

This PR adds training and configuration scripts for a new supervised fine-tuning (SFT) run targeting the "gim-sft-2" experiment. The implementation sets up model training using the Unsloth framework with Qwen3-4B-Instruct as the base model.

• Implements complete training pipeline with model loading, dataset preparation, and training configuration
• Adds configuration management for experiment parameters including batch sizes, learning rates, and data splits
• Excludes the runs directory from mypy type checking

Reviewed Changes

Copilot reviewed 3 out of 6 changed files in this pull request and generated 2 comments.

File	Description
runs/09191-gim-sft-2/train.py	Main training script implementing the complete ML pipeline from model loading to inference testing
runs/09191-gim-sft-2/configs.py	Configuration module defining experiment parameters and directory structures
.pre-commit-config.yaml	Excludes runs directory from mypy type checking

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}