Set up Cursor Cloud dev environment (AGENTS.md + startup deps)

[ShaneHurley]

What

Sets up the development environment for this NBA/Basketball ELO notebook repository so cloud agents can run the notebooks in Jupyter.

• Adds AGENTS.md documenting the project layout and durable, non-obvious run/setup caveats under ## Cursor Cloud specific instructions.
• Configures a minimal, idempotent startup update script (via the environment setup tool) that installs the full scientific + NBA stack and Jupyter: pandas==2.2.3 numpy scipy scikit-learn xgboost catboost optuna tqdm nba_api nbainjuries matplotlib requests jupyter notebook ipykernel nbconvert jupyterlab (using pip install --break-system-packages).

No existing code was modified.

Why these notes matter

• Notebooks were authored on Google Colab and read CSVs from a mounted Google Drive (from google.colab import drive, paths like /drive/MyDrive/basketballData/...). Those cells fail here and must be skipped/repointed — data is not in the repo.
• nba_api / nbainjuries rely on external network (stats.nba.com), which may be slow or blocked in the sandbox.
• Jupyter CLI scripts live in ~/.local/bin (not on PATH); run via python3 -m jupyter ....

Verification

• All 6 notebooks validate (nbconvert --to script).
• Launched JupyterLab dev server and ran a hello-world notebook exercising the project's core ELO update algorithm (the calc formula from NBA_ELO.ipynb) on a synthetic possession simulation: single-update direction check passes, and time-averaged offensive ELO recovers the hidden true offensive ordering (Spearman = 1.000).

jupyterlab_elo_demo_run.mp4

ELO results table and Spearman = 1.000 PASS
Core ELO algorithm separates strong vs weak offenses

_{To show artifacts inline, enable in settings.}