Shinka Setup

Scaffolds a new ShinkaEvolve task by generating `evaluate.py` and `initial.<ext>` from your task description. You describe what you want to optimize, pick a language (Python, Rust, Julia, C++, etc.), and it writes both the baseline solution with evolve markers and the evaluation harness that scores candidates. For Python it uses `run_shinka_eval`, for other languages it shells out and writes JSON. It runs a smoke test to confirm the evaluator actually works before handing off, which saves you from debugging broken setups mid-evolution. Use this when you want to throw an optimization problem at LLM-driven evolution but don't want to write the boilerplate yourself. It'll optionally stub out `run_evo.py` and `shinka.yaml` if you want to kick off a run immediately.