Summary

This PR fixes issue #5247 where codebase indexing was performing incorrect code parsing, creating arbitrary code fragments instead of complete function and class definitions.

Problem

The tree-sitter parser was generating chunks containing only 3-4 lines of random code (like try-catch blocks or simple assignments) rather than complete functions or classes. This was happening because the parser logic was incorrectly processing tree-sitter captures.

Solution

Fixed the implementation in to:

• Properly use tree-sitter captures to identify complete function and class definitions
• Process captures to find definition boundaries and create chunks for entire functions/classes
• Add logic to avoid duplicate processing of the same definition nodes
• Maintain fallback chunking only when no valid definitions are found

Key Changes

• Modified the capture processing logic (lines 125-200) to correctly identify definition nodes from captures
• Added proper handling for both direct definition captures (, ) and name captures (, etc.)
• Ensured chunks contain complete code definitions rather than arbitrary fragments
• Added deduplication logic to prevent processing the same definition multiple times

Testing

• Verified the fix works correctly with Python code containing functions and classes
• Confirmed that complete definitions are now captured as single chunks
• All existing tests pass and type checks are clean

Fixes #5247