Add mission-ready GMAT interop tests: 3 expected failures
Tests that expose real gaps in our libraries when compared against GMAT:
ocaml-odm (2 FAIL):
- LEO interpolation midpoint: linear vs quadratic truth returns inf —
exposes precision bug in interpolate epoch conversion
- Molniya perigee interpolation: same issue at high-eccentricity perigee
ocaml-collision (1 FAIL):
- TCA precision: miss distance varies by 34 km in 10s at closest approach.
Step-based TCA finding is not mission-grade — need quadratic refinement.
ocaml-cam (new mission-ready tests, passing):
- Linear model (dv*dt) vs GMAT vis-viva SMA prediction (< 5% error)
- CAM evaluate internal consistency check
- Documents that linear along-track shift model is a first-order
approximation only — GMAT shows the real effect is an SMA change.
These failures are intentional — they document what needs to be fixed
before the libraries are mission-ready.