Fix OpenML checkpoint directory and filenames

[charlesmartin14]

Motivation

• The notebook created an extra intermediate runs folder and embedded the OpenML id in every checkpoint filename, making directories and filenames verbose and brittle.
• The intent is to keep the dataset/model id in the folder name but use stable, id-agnostic filenames inside the folder so checkpoints are easier to locate and manage.

Description

• Remove the intermediate RUN_NAME_BASE folder and set RUNS_ROOT = Path('/content/drive/MyDrive/xgbww_runs') so runs are stored directly under that root.
• Change per-dataset checkpoint folder naming to openml_<dataset_id>_w1w2w7w8w9 (removed the _alpha suffix).
• Replace per-file names that included the dataset id with generic checkpoint filenames: state.json, metrics.csv, model_latest.json, data_split.npz, summary.json, and model_meta.json.
• Update the notebook logic that writes model_meta to reference the new runs_root, checkpoint_dir, and the new checkpoint_files names.

Testing

• Ran python -m json.tool XGBWW_OpenML_W1W2W7W8W9_for_openml_id.ipynb to validate the notebook JSON, which succeeded.
• Verified the updated path and filename patterns with rg searches for the new RUNS_ROOT, DRIVE_ROOT, and checkpoint path variables, which matched the expected replacements.
• Confirmed the notebook cell that derives dataset_id still populates the folder name so random or user-specified ids remain present in the directory name.

---

Codex Task