NASA-IMPACT/SuryaBench — reverse-engineered prompt

Build me a clean, usable benchmark repo called SuryaBench for testing AI models on heliophysics and space weather tasks. I want it centered on dataset prep and evaluation for six applications, active region segmentation, active region emergence prediction, coronal field extrapolation, solar flare forecasting, solar wind forecasting, and solar EUV spectra modeling. The main thing is that each task should have the code and instructions needed to generate or organize its dataset, with a consistent layout and simple examples so someone can actually run it without digging through everything.

Please wire it up so the project can use the published SuryaBench data on Hugging Face when available, and otherwise provide the task specific scripts for creating labels, splits, and inputs described in the docs. Add a top level README that explains what each task is in plain English, how to get the data, and how to run each pipeline. Keep the Python side straightforward, and if any task needs a separate compiled component, document that clearly. You can look up current docs online if needed.