Fix for Gemini 2.5 Pro and improve dependency handling

[szmania]

Summary

This pull request introduces support for Google's Gemini 2.5 Pro model and other related Gemini models. Key improvements include adding new model aliases, enhancing environment validation to check for the
GEMINI_API_KEY, and implementing a user-friendly mechanism to detect and install required Python packages automatically.

Key Changes

1 Model Aliases and Configuration:
• Added new aliases in aider/models.py to easily access Gemini models, including:
• gemini-2.5-pro
• gemini (aliased to gemini/gemini-3-pro-preview)
• flash (aliased to gemini/gemini-2.5-flash)
• Updated the model validation logic to recognize and require the GEMINI_API_KEY environment variable when a Gemini model is selected.
2 Automatic Dependency Installation:
• To provide a smoother user experience, we've added a check that verifies if the necessary client library (google-generativeai) is installed when using Gemini models.
• If the package is missing, the system now raises a specific, informative error and prompts the user to install it automatically via pip, as seen in the new logic within aider/exceptions.py.
3 Enhanced Exception Handling:
• The LiteLLMExceptions class in aider/exceptions.py was updated to catch APIConnectionError related to google.auth issues. This allows us to provide a clear, actionable error message to the user instead of a
generic connection error.
4 Updated Tests:
• Added new asynchronous tests in tests/basic/test_models.py to validate the new dependency-checking functionality, ensuring the user is correctly prompted to install missing packages for various model providers
like Google (Vertex AI) and AWS (Bedrock).