Document ellipsoid ode45 source-sampling gap

[cmccomb]

What changed

• Capture both pinned MATLAB Nonlinear_Hydro/ode45 baselines, Regular and RegularCIC, with saved body, PTO, wave, and force histories.
• Add source-force diagnostics and bounded Python trajectory comparisons for those cases, while retaining the strict ode4 parity gates.
• Include all three ellipsoid selectors in the routine MATLAB reference-model matrix and document the remaining ode45 gap.

Finding

The physical inputs are identical to the published ode4 cases except for the MATLAB solver. In both ode45 runs, logged restoring force at sample i equals mesh buoyancy evaluated at body state i-1 within 6e-9 N, while its current-state mismatch exceeds 54 kN. The logged total force includes this lagged term, and the adjusted Simscape mass times logged acceleration agrees with total plus PTO force within 2e-9 N. Drag and excitation remain current-state forces. The source's ode4 and ode45 trajectories differ by up to 15.9 mm heave and 35.4 kN PTO force.

Python continues to apply instantaneous buoyancy. Its ode45-source comparisons are within 19.94 mm heave and 36.05 kN PTO force over 150 s, but this is recorded as a solver-envelope diagnostic, not ode45 numerical parity. No delay or coefficient tuning was added to Python.

Verification

• Fresh pinned R2025b ode45 Regular and RegularCIC baselines completed successfully.
• Local paired diagnostics: 4 passed for ode45; prior ode4 Regular and RegularCIC gates: 2 passed each.
• Local suite: 90 passed, 93 artifact-dependent tests skipped.
• actionlint and git diff --check passed.

SungJun Won's original commits remain in branch ancestry.