graphcast: handle static vars without a time dim (GFS path)

Add GFS init source as alternative to ERA5T (Phase 2 wiring)

Cut dead neural/LSTM pricing + predictor paths

Replace neural actuarial model with GraphCast triggers + empirical burn analysis

Fix circular import between heat_forecast and lstm_model

Add LSTM neural model, ERA5 data, FAISS+BM25 RAG, Neon DB, eval suite; de-jargon frontend

Initial commit: Extreme Heat Risk Engine