This pull request introduces several improvements and updates across the codebase, primarily focusing on enhanced support for DeepSeek models, improved file encoding practices, and better cross-platform compatibility for file locking. The most significant changes are grouped below:

1. DeepSeek Model Support and Routing

• Added DeepSeek models (openai/deepseek-v4-flash, openai/deepseek-v4-pro) to the SUGGESTED_MODELS list in model_switcher.py, making them available for selection.
• Updated the logic in _should_replay_thinking_state to recognize DeepSeek models in addition to Anthropic, ensuring correct handling of replayed thinking metadata.

2. File Encoding Consistency

• Standardized all file read/write operations to use UTF-8 encoding in the following files and functions:
  • manager.py (_load_system_prompt)
  • session.py (save_trajectory_local, update_local_save_status) [1] [2]
  • session_uploader.py (_write_row_payload, _write_claude_code_payload, _read_session_file, _update_upload_status) [1] [2] [3] [4]

3. Cross-Platform File Locking

• Improved file locking logic in session_uploader.py to check for the presence of fcntl (which is unavailable on Windows), allowing the code to run on more platforms without errors. [1] [2] [3]

4. Testing

• Added a new test in test_litellm.py that loads environment variables and runs a completion using the DeepSeek chat model, verifying integration.