Add MLIPX PaiNN-style MLIP project: data pipeline, model, training, inference, and tests

[yeze5421]

Motivation

• Provide a minimal research/engineering codebase for training PaiNN-style atomistic potentials with ASE/extxyz datasets and energy+force supervision.
• Replace the prior toy example with a structured, testable package that supports training, evaluation, inference, and an ASE Calculator wrapper.
• Offer reproducible defaults and convenience scripts to create demo data, run training/inference/evaluation, and package the project.

Description

• Add package layout and metadata including pyproject.toml, README.md, and two config files configs/minimal.yaml and configs/standard.yaml with sensible defaults.
• Implement configuration utilities in src/mlipx/config.py and data handling with ASEDataset, MLIPDataModule, neighbor graph builder, and GraphBatch types under src/mlipx/data/.
• Implement model components: RBF and scatter utilities in models/layers.py, a PaiNN-style message-passing network in models/painn.py, and a high-level MLIPEnergyModel with predict_energy_forces in models/model.py.
• Add training engine mlipx.training.engine.Trainer that handles optimization, LR scheduling, AMP, checkpointing, early stopping, and a CLI entrypoint src/mlipx/train.py plus evaluate.py and infer.py for evaluation and inference workflows.
• Provide ASE integration with src/mlipx/ase_ext/calculator.py exposing build_calculator, convenience scripts under scripts/ (train.sh, evaluate.sh, infer.sh, make_demo_dataset.py, package.sh), project __init__, and utility modules for metrics and seeding.
• Remove the old monolithic mlip.py teaching demo and add unit tests in tests/ to cover the data pipeline and a smoke training/model forward pass.

Testing

• Added unit tests tests/test_data_pipeline.py and tests/test_model_and_train.py that exercise dataset reading, graph collate, model forward and force prediction, and a one-epoch smoke training run using a temporary extxyz dataset.
• Ran the test suite with pytest -q and the tests completed successfully.

---

Codex Task