[Improvement] Enhance engine error handling and documentation add more logging and doc

[louiswang524]

Summary

• Improve error handling in Engine initialization
• Add better error messages for debugging
• Fix page_size configuration usage
• Add documentation for helper functions

Changes

1. Error Handling Improvements:

   • Replace assert not torch.cuda.is_initialized() with proper RuntimeError exception
   • Replace assert for insufficient memory with detailed ValueError message including:
     • Available memory information
     • Memory per page calculation
     • Suggestions for configuration adjustments

2. Configuration Fix:

   • Use config.page_size instead of hardcoded 1 in Context initialization (line 79)
   • Note: page_size=1 is still the only supported value (enforced in core.py), but this makes the code consistent

3. Documentation:

   • Add docstring to _align_up_32() explaining 128-byte alignment purpose
   • Add return type hint to Engine.__init__()

Testing

• Existing tests should pass
• Better error messages will help users diagnose memory issues