Declarative study configuration from YAML/TOML

[jc-macdonald]

Problem

Studies can only be defined in Python. Non-programmers or CI pipelines would benefit from a declarative config format.

Proposed

• Study.from_yaml(path) / Study.from_toml(path) class methods
• Maps factor definitions, observables, phases, and grid methods to a config file
• Simulator and scorer still referenced by importable dotted path (e.g., my_module.MySimulator)

Example

observables:
  - name: rmse
    direction: minimize
  - name: train_time
    direction: minimize

factors:
  - name: n_estimators
    type: discrete
    levels: [50, 100, 200]

phases:
  - name: discovery
    method: lhs
    n_samples: 100
    filter: top_k_pareto(20)

Notes

Low priority — Python API is primary. This is a convenience layer.