[sylvesterkaczmarek]

For nonuniform sample depths, volume_rendering reuses the first gap at the far endpoint. For depths [0, 1, 4], this produces intervals [1, 2, 2] instead of [1, 2, 3], changing the final sample's opacity, rendered values, and depth.

Use the adjacent final gap instead. Update the existing test fixture and add independent numerical regressions covering increasing/decreasing spacing, uniform and two-sample controls, batched rays, JIT, and opaque_final_sample.

Validation: python -m pytest -q jax3d/math/volume_rendering_test.py passes all 71 tests. Eight nonuniform regression cases fail on unchanged upstream; the eight control cases already pass. Syntax compilation and git diff --check also pass.

[sylvesterkaczmarek]

Any feedback on the volume-rendering endpoint correction? The workflow is still waiting for maintainer approval.