Save trained FedCA checkpoint and add standalone test.py

[hongleizhang]

Motivation

• Persist the full personalized federated state after training so evaluations and downstream use can reload exact model and client parameters.
• Provide a standalone evaluation entry point that can load a saved checkpoint and compute test metrics without retraining.
• Document checkpoint/save-and-evaluate workflow for reproducible experiments.

Description

• Added Engine.saveCheckpoint and Engine.loadCheckpoint to model/engine.py to save and restore config, model_state_dict, client_model_params, server_model_param, and agg_participant_index_map.
• Added a --model_path argument to train.py and call engine.saveCheckpoint(config['model_path']) after training to write the checkpoint file (default checkpoints/fedca.pt).
• Added a new standalone test.py which loads the checkpoint, restores engine state, rebuilds dataset splits with SampleGenerator, and runs engine.federatedEvaluate to report HR and NDCG (supports --use_cuda, --device_id, and --top_k).
• Updated README.md with examples for saving a checkpoint and running the independent evaluator (python train.py --model_path=... and python test.py --model_path=...).

Testing

• Ran git diff --check, which passed without issues.
• Byte-compiled train.py, test.py, and model/engine.py with python -m py_compile, which succeeded.
• Tried python test.py --help in this environment but it failed to import torch (ModuleNotFoundError: No module named 'torch'), so end-to-end loading/evaluation was not executed here.
• Changes were committed and the working tree is clean after the commit.

---

Codex Task