Pure Rust Support for MLE-Bench / Custom Task Bootstrap (Reduce Python Dependency)

[micahstubbs]

Goal: Reduce or eliminate the need for Python-side scripts when setting up MLE-Bench or custom tasks in sia_rust.

Part of the broader effort to deprecate the Python bridge (#137) and move toward a more self-contained Rust experience.

Tasks

• Port or re-implement the MLE-Bench dataset download/prep logic in Rust (or provide a clean one-command native path).
• Make task discovery and evaluate contract work without requiring a Python environment for the core sia run flow.
• Keep Python evaluate.py support during transition but mark as deprecated for new tasks.

Related: #137 (deprecation), #20 (original task system), #99 (new legal benchmark task).