Add internal canonical gains v2 builder

[uriahf]

Summary

Adds the smallest internal-only canonical rtichoke_viz v2 builder for gains.

• reuses the semantic evaluation metadata and deterministic evaluation/series identity pattern established by the ROC v2 builder
• emits gains data as ppcr × sensitivity
• keeps the random benchmark global
• emits one population-owned perfect-model path per semantic population
• derives prevalence from existing production real_positives / n quantities rather than recalculating outcomes in the visualization layer
• maps compatibility/reference groups through semantic metadata before assigning population ownership
• keeps equal-prevalence populations as distinct reference owners
• preserves model-unknown population semantics
• does not switch production rendering to v2
• does not change public APIs or statistical calculations

Tests

Adds focused coverage for:

• prevalence-dependent perfect path geometry
• two models sharing one population and one perfect reference
• equal-prevalence populations remaining distinct owners
• existing ROC-v2 behavior remains covered

CI should verify formatting, typing, and the full Python test suite.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-08-22 13:05 UTC