Implement Runner module: grid execution and adaptive search

[jc-macdonald]

Summary

Port runner.py for parallel grid execution and adaptive multi-objective search.

Functions to Implement

Function	Description	Julia approach
run_grid(model, grid; parallel=true)	Execute model over design matrix	Threads.@threads or pmap
run_adaptive(model, factors, scorer; n_trials, objectives)	Multi-objective adaptive search	Metaheuristics.jl NSGA-II

Design Notes

• Parallelism: Julia has native Distributed/Threads — no joblib equivalent needed. Use Threads.@threads for shared-memory, pmap for distributed. Let user pass a parallel_backend kwarg.
• Adaptive search: Metaheuristics.jl provides NSGA-II/III for multi-objective optimization, matching Optuna's NSGA-II sampler. Alternatively, Surrogates.jl provides EI+LCBS surrogate-based optimization for expensive objectives.
• The runner should accept any AbstractModelWorld and return a ResultsTable.

Acceptance Criteria

• run_grid produces correct number of result rows
• run_adaptive improves over random search on a known test problem
• Both functions return ResultsTable conforming to the Protocols module
• Thread-safety tested