misc: split ps.py file into multiple files

[specture724]

Most of the functions and classes are in ps.py, making it so large. We need to split it.

• Move code related to RDMA devices and P2PStore into p2p_store.py
• Move all data structures into data_types.py
• Move code related to read files and register pin memory into pin_memory.py
• Move fastapi code into api.py
• Move run_from_cli to __main__.py
• export all public interfaces in __init__.py

[Copilot]

Pull request overview

This PR refactors the large ps.py file into multiple smaller, more focused modules to improve code organization and maintainability. The core functionality remains unchanged; the refactoring only reorganizes existing code into logical units based on their responsibilities.

Key changes:

• Extracted RDMA device handling and P2P store functionality into p2p_store.py
• Moved all Pydantic data models and type definitions into data_types.py
• Separated checkpoint loading and pin memory operations into pin_memory.py
• Isolated FastAPI endpoints and HTTP communication into api.py
• Created __main__.py as the CLI entry point
• Updated __init__.py to expose all public interfaces

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
checkpoint_engine/data_types.py	New file containing all Pydantic models and type definitions (ParameterMeta, BucketRange, H2DBucket, etc.) with validators and serializers
checkpoint_engine/p2p_store.py	New file with RDMA device detection/parsing functions and the P2PStore class for managing peer-to-peer memory transfers
checkpoint_engine/pin_memory.py	New file with checkpoint loading functions (_load_checkpoint_file, _register_checkpoint) and memory pinning operations
checkpoint_engine/api.py	New file containing FastAPI application initialization (_init_api) and HTTP endpoint definitions for checkpoint management
checkpoint_engine/__main__.py	New file with CLI entry point (run_from_cli) for starting the parameter server via uvicorn
checkpoint_engine/__init__.py	Updated to export all public interfaces from the new modules with explicit all list
checkpoint_engine/ps.py	Significantly reduced file that now imports from the new modules and focuses solely on ParameterServer class implementation
tests/test_rdma_parser.py	Updated imports and patch paths to reference the new checkpoint_engine.p2p_store module location

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[blahgeek]

merged in #75