feat: Add diff hunk parsing with line-by-line change extraction

[nick-galluzzo]

Enhances git diff parsing to extract actual code changes at the line level, providing structured data optimized for AI processing. This foundational improvement enables more intelligent commit message generation by giving AI models the actual changed content rather than just metadata.

Key Changes

Enhanced Data Models

• Added DiffHunk and HunkLine classes for structured diff representation
• Enhanced FileDiff with hunks field containing actual change content
• Added utility methods for accessing added/removed content

Improved Diff Parser

• Modified GitDiffParser to extract line-by-line changes using unidiff
• Preserved existing functionality while adding detailed hunk parsing
• Added support for hunk headers and section context

AI-Optimized Output

• Added get_ai_context() method that outputs minimal git diff format
• Research-backed approach focusing on essential changes without git plumbing metadata
• Format optimized for LLM training data alignment

Technical Details

• Uses unidiff library for robust diff parsing instead of custom regex
• Maintains backward compatibility with existing CommitAnalysis.to_ai_context()
• Clean separation between structured data (for analysis) and AI format (for generation)

Technical Decisions

• The get_ai_context() method returns minimal git diff format (matching git diff output) based on research showing LLMs perform best with familiar training data formats
• Chose unidiff library over custom parsing for reliability and maintainability. Unidiff allows us to break apart each hunk and better categorize and data engineer our git diff data components.
• MVP approach prioritizes proven formats over custom optimizations

Testing

• Manual testing with various staged changes (add/modify/delete/rename)
• Verified output matches standard git diff format exactly