fix: Suppress harmless multiprocessing warnings from ydata-profiling

fix: Support follow-up requests without re-uploading file

fix: Configure high-quality ydata profiling for HuggingFace 16GB RAM

feat: Add dynamic prompt system for small context window models (Groq support)

fix: Use stable gemini-2.5-flash instead of experimental model with stricter rate limits

fix: Support both GOOGLE_API_KEY and GEMINI_API_KEY env vars for HuggingFace compatibility

fix: Auto-optimize ydata profiling for memory constraints

fix: Clarify report intent and fix parameter issues

fix: Handle output_dir parameter mismatch in ydata_profiling

fix: Correct ToolCall object structure for Gemini text parser

feat: Implement text-based tool calling for Gemini (ReAct pattern)

fix: Remove system_instruction parameter from chat endpoint

fix: Add Gemini safety settings and handle blocked responses

fix: Remove Gemini tool calling - use text-only mode

fix: Remove invalid parameter from start_chat()

fix: Configure Gemini tools at model level, not send_message level

fix: Resolve duplicate 'tools' argument in Gemini API calls

fix: Fix module import paths for Render deployment

fix: Resolve blank modal issue when viewing YData Profiling reports

fix: Remove all remaining Sweetviz references

fix: Add robust error handling for sessions database initialization

refactor: Remove Sweetviz and use YData Profiling as primary EDA tool

fix: Add NumPy 2.x compatibility handling for Sweetviz

feat: Initial commit - Data Science Agent with React frontend and FastAPI backend