Fix CVT projectors for aligned SVT and BMT geometry

[yugotra]

Problem

At 5 T, the numerical CVT Kalman projector returned perturbed helix states to simplified detector surfaces. For aligned SVT sensors it intersected only the plane trace in x-y and discarded the plane normal's z component. For aligned BMT layers it intersected a cylinder centered on the beam axis instead of the displaced and tilted geometry axis. The projector therefore differentiated a different measurement geometry from the geometry used by normal state transport.

This explains why the main SVT failure was absent for ideal-geometry MC and 0 T data: ideal SVT planes have no omitted longitudinal tilt, while the 0 T straight-track path already uses the full plane intersection.

Change

• Intersect analytic helices with the complete SVT plane, including its longitudinal tilt.
• Intersect analytic helices with the true BMT cylinder axis.
• Preserve the exact closed-form cylinder result for ideal beam-axis geometry.
• Use an analytic Newton derivative with a bounded bracket-and-bisection fallback for the cylinder solve.
• If either finite-difference perturbation cannot reach its aligned cylinder, mark that measurement unusable instead of substituting the nominal beam-axis crossing.
• Add eight unit tests covering tilted planes, displaced and tilted cylinders, both helix signs, ideal-geometry identity, the bracket fallback, and a degenerate axis.

The state still swims through the field map while the projector uses an analytic uniform-field helix. This change corrects the detector-surface geometry used by that analytic projector; it does not make the projector field-map exact. The swimmer-based projector was tested and performed worse because finite differences amplify swimmer noise.

Validation

• HelixTest: 8 tests passed.
• Full 49-module Maven package build passed.
• SVT-only 5 T A/B: valid filtered covariance 31.00% -> 99.73%; residual robust width 52.83 -> 14.89 um.
• Full production CVT, run 19060, all 556,309 events, beamSpotConst=2: valid covariance 87.39% -> 94.93%; SVT residual width 109.76 -> 54.89 um; layerwise width cos(2phi) amplitude 29.8-66.5 -> 0.14-1.91 um.
• SVT controls: ideal MC and 0 T outputs were byte-identical; 4.07 million plane solves had no non-convergence.
• BMT 20k A/B: residual width 287.95 -> 245.10 um and median chi2/NDF 1.85 -> 1.74. The valid-covariance fraction was unchanged, so this is an independent BMT residual improvement.
• BMT controls: ideal MC and 0 T outputs were byte-identical. A 100,000-case BMT-like solver audit found no failures, wrong residuals, or root jumps.

The unrelated covariance-reporting correction and the broader swimmer-recovery policy change are intentionally excluded from this branch.

[raffaelladevita]

Currently, the track propagation in the magnetic field is done with the swimmer. Therefore the changes to the helix-based propagation would have no effect. The change that may be critical is the check on the matrix H.

What problem was found with the swimmer?