Fix RSPHERE particle-boundary momentum basis

[tomzhu0225]

Summary

RSPHERE stores phi as elevation from the Cartesian x-y plane. The particle-boundary kernel instead used the spherical basis for a polar angle measured from +z. Consequently, reflecting a particle at a radial boundary changed the wrong Cartesian momentum components away from special angles.

This change:

• uses the documented elevation-angle basis for Cartesian/local momentum conversion;
• samples thermal re-emission in the local (u_r, u_theta, u_phi) frame, so the half-Maxwellian normal component is always directed into the domain;
• adds native RSPHERE reflecting and thermal regressions with 32 angularly distributed particles.

Reproducer and validation

On the unmodified development branch:

• reflecting boundary: 96/96 Cartesian momentum-component checks fail;
• thermal boundary: 8/32 returned particles point outward.

With this commit:

• reflecting run and component-wise analysis pass;
• thermal run and inward-normal analysis pass;
• Release RSPHERE build, Ruff, Python compilation, repository input-name audit, and git diff --check pass.

This is independent of #7194, which corrects Cartesian-to-RSPHERE particle position initialization. These tests use native NUniformPerCell RSPHERE initialization and directly exercise the boundary kernel.