[jacksteamdev]

Summary

Move numerical regression ownership out of website deployment and into the library.

normal library CI → True Sky source fixture
explicit full-data verification
  → historical Fagan–Bradley native + Wasm replay (separate dataset input)
  → production dataset segment parity

Fixtures are preserved byte-for-byte. Historical numeric comparisons allow 1e-12 rounding while retaining exact structure and ordering. No runtime behavior changes.

Evidence

• Before: Website release v0.2.0 failed on a 7–14e-15 degree cusp difference in an exact legacy comparison.
  After: All eight preserved historical cases pass in native Rust and Wasm against dataset 06d4c26e…; negative controls reject larger changes, reordered bodies and array/object substitutions.
• True Sky fixture, Rust formatting/Clippy and browser-example TypeScript checks pass. Independent review's structural comparison finding is fixed.
• Full-data workflow remains explicit; its production segment invocation was wired but has not been run against the distinct production dataset in this change.

Merge Danger

Door: two-way

Blast Radius: verification

Adds tests and an optional full-data workflow input. Website cleanup follows this PR; no application release is triggered.