Add height field support to mjx.ray

[elliot-at-liminalnook]

mjx.ray has no intersection function for HFIELD geoms, so the dispatch loop silently skips height fields: every ray that should hit terrain instead returns (dist=-1, geomid=-1) while the same ray hits with mujoco.mj_ray. Issue #2155 tracked this and was closed once the Warp backend gained hfield rays, but the JAX backend still lacks it.

This adds _ray_hfield, a port of mj_rayHfield: the base box, the two surface triangles per grid cell, and the four prism side faces, where a side hit only counts if the hit point lies below the terrain edge interpolated between the neighboring data points. It also makes ray warn when a candidate geom has a type with no registered ray function, instead of silently treating it as a miss, and updates the one-line MJX feature-support entry.

I want to be upfront that there is an open in-flight PR for the same feature, #2804, which I only found after writing this. I'm posting this as complementary rather than competing. The overlap is the core port; what this adds on top is the silent-miss warning for unsupported geom types, an exact-parity test battery (9 rays checked bitwise against mj_ray: straight-down and oblique surface hits, side-face hits below the terrain edge, the base box from below, occlusion by a sphere above the terrain, and misses), and the doc-table line. The implementations differ in shape: this one composes per-face candidate distances with a running minimum and a per-face validity mask, whereas #2804 uses a compiled-for-fixed-size early-out. I'm happy to converge on whichever implementation the maintainers prefer, drop mine in favor of #2804's, or fold these tests and the warning onto #2804; the tests are implementation-agnostic and should transfer either way.

Separately, while testing I hit a pre-existing crash unrelated to this change: mjx.ray raises when the model has no materials (nmat == 0), because the visibility filter indexes mat_rgba[geom_matid] with geom_matid == -1 into an empty array. All the tests here put a material on every geom to avoid it. I'm happy to send a follow-up that guards that filter if useful.

I prepared this change with AI assistance (Claude); I have not added it as a commit co-author, since it cannot sign the CLA, and I've reviewed the change and take responsibility for it.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.