Adds Windows support

[ramkrishna2910]

Windows Support for Human-Eval

This PR adds comprehensive Windows support to the code evaluation pipeline. Here are the key changes:

Process Management

• Replaced Unix-specific Process handling with Windows-compatible Pool implementation
• Added proper process cleanup using pool.terminate() and pool.join() to prevent zombie processes on timeout
• Switched to 'spawn' context instead of 'fork' for Windows compatibility: multiprocessing.get_context('spawn')

Timeout Implementation

• Added Windows-specific timeout mechanism using threading since SIGALRM is not available on Windows
• Implemented thread-based timeout using PyThreadState_SetAsyncExc to inject exceptions into running threads
• Added proper thread state cleanup to prevent memory leaks and ensure thread state consistency
• Introduced done and raised events to handle race conditions in timeout scenarios

Memory Management

• Added Windows-specific memory limits using Windows Job Objects
• Integrated pywin32 package for Job Object API access
• Added conditional dependency: pywin32>=306; platform_system == "Windows"
• Implemented memory limits through JOB_OBJECT_LIMIT_PROCESS_MEMORY flag

Platform Detection

• Added IS_WINDOWS flag using platform.system() for conditional code paths
• Separated Unix and Windows-specific implementations into distinct functions
• Made all path handling Windows-compatible

These changes ensure that the human-eval framework runs reliably on Windows while maintaining feature parity with Unix systems. The implementation handles Windows-specific quirks around process management, timeouts, and memory limits while preserving the security and isolation guarantees of the original implementation.

Following is the sanity check using the example results done on Windows:

Closes #51 #45 CC #50