Implement Study module: multi-phase orchestration

[jc-macdonald]

Summary

Port study.py for multi-phase simulation study orchestration with carry-forward.

Types and Functions to Implement

struct Phase
    name::String
    factors::Vector{Factor}
    design_method::Symbol  # :full, :lhs, :adaptive
    n_trials::Int
    scorer::AbstractScorer
    filter::Function       # e.g. top_k_pareto_filter
end

struct Study
    phases::Vector{Phase}
end

Function	Description
run!(study, model)	Execute all phases sequentially, carry-forward winners
top_k_pareto_filter(results, k; objectives)	Keep top-k Pareto-optimal configurations

Design Notes

• Phase N's output feeds Phase N+1's input: winners from Phase 1 screening become the reduced factor set for Phase 2's grid, etc.
• Carry-forward pattern: each Phase stores its ResultsTable; the Study orchestrator feeds phase results into the next phase's factor reduction.
• The Python version has 3 tiers: Screen → Grid → Adaptive. Julia should keep the same pattern but make it generic.

Acceptance Criteria

• Two-phase study (screen → grid) executes end-to-end
• Carry-forward correctly reduces factor set
• Phase results accessible after study completion