Introduce ma_reap with agent assignments, stakes, learned CV weights, and sidecar outputs; add scipy dependency and bump version to 0.2.0.