Add more examples for few-shot learning

[ishikawa]

This PR enhances the few-shot learning capabilities of our project by adding more example outputs and implementing a parsing mechanism for these examples.

Key changes:

• Added new example PR outputs in Markdown format (sum_diff/examples/001.md, 002.md, and 003.md)
• Implemented parse_pr_example function in sum_diff/utils.py to process example PR outputs
• Updated main function in sum_diff/__init__.py to incorporate example outputs into the user prompt
• Added unit tests for the parse_pr_example function in tests/test_utils.py
• Updated pyproject.toml to include pytest configuration and dev dependencies

The new examples provide a diverse set of PR formats, including different styles and complexities. This will help improve the quality and variety of generated PR titles and descriptions.

The parse_pr_example function extracts the title and description from the Markdown-formatted example files, handling HTML comments and multiple headings. This parsed data is then incorporated into the user prompt, allowing for more effective few-shot learning.

Example usage of the new parse_pr_example function:

from sum_diff.utils import parse_pr_example

markdown_content = """
<!-- This is a comment -->
# Example PR Title

This is the PR description.
"""

pr_example = parse_pr_example(markdown_content)
print(pr_example.title)  # Output: Example PR Title
print(pr_example.description)  # Output: This is the PR description.

These changes will enhance the AI model's ability to generate more accurate and contextually appropriate PR titles and descriptions based on the provided examples.