v0.1.0

DecisionRules v0.1.0

Closed issues:

• Implement per-stage decomposition with DiffOpt implicit differentiation and objective sensitivity (#6)
• Extend Deterministic Equivalent policies to work with general uncertainties (#15)
• Target-penalty annealing: a constant slack-penalty weight either hides target violations or degrades the learned policy (#41)
• Deterministic-equivalent evaluation can overstate policy quality — add a stage-wise rollout evaluation and a target-violation metric (#42)