@hackage shikumi-eval0.2.0.3

The evaluation framework for shikumi: the owned data model (Example, Prediction, Dataset, Metric, Score, Report — MasterPlan integration point #5), built-in pure and LM-backed metrics, an evaluate runner that scores a Shikumi.Program.Program over a typed dataset with bounded parallelism and per-example error boundaries, and golden testing that pins a program's behaviour deterministically under a mock or replayed LM.