Add code search tool with RAG capabilities

[neubig]

Description

This PR adds a new code search tool that uses Retrieval Augmented Generation (RAG) to enable semantic code search across repositories.

Features

• Semantic code search using sentence transformers (configurable via env var)
• Fast similarity search using FAISS
• Support for indexing any git repository with configurable file extensions
• Save/load functionality for search indices
• Comprehensive error handling and test coverage

Implementation Details

• Added new dependencies: PyTorch CPU, Sentence Transformers, FAISS
• New module structure:
  • code_search/core.py: Core indexing and search functionality
  • code_search/tools.py: High-level tool functions
  • Tests in tests/test_code_search.py

Example Usage

# Initialize search for a repository
result = initialize_code_search(
    repo_path="/path/to/repo",
    save_dir="/path/to/save",
    extensions=[".py"],  # optional
    embedding_model="BAAI/bge-base-en-v1.5"  # optional
)

# Search code
result = search_code(
    save_dir="/path/to/save",
    query="function that handles HTTP requests",
    k=5  # number of results
)

Testing

• Added comprehensive unit tests
• Tested on the openhands-aci repository itself with good results
• Example search results for "code that handles file editing":

  File: openhands_aci/editor/__init__.py
  Score: 0.727
  ...
  

Notes

• The embedding model can be configured through an environment variable
• All tests are passing
• Documentation included in code