[jeffamstutz]

Summary

ANARI 1.1 defines near/far clip distances on perspective and orthographic cameras, but VisRTX ignored them. The spec also never says what channel.depth holds for a pixel with no hit, and the RTX renderers disagreed on it:

• fast, interactive and default left FLT_MAX.
• debug and quality wrote the primary ray's t.upper. A cutting plane could shorten that to the cut distance, or to 0.

This PR makes them consistent:

• Clip planes: cameras read near/far as clip planes measured from position along direction, and every primary ray is bounded by them. Surfaces and volumes outside the range are clipped.
• Miss Depth: a pixel with no hit reports its ray's Far Limit:
  • far / cosθ for perspective;
  • exactly far for orthographic;
  • ray.tmax for the ray-buffer camera;
  • FLT_MAX when nothing bounds the ray, which matches what apps see today by default.
• Cutting planes no longer change what a miss reports.
• Invalid ray buffer: it now misses with an empty range at FLT_MAX instead of [0, 0].
• Invalid near/far: a pair that breaks the spec's rules (0 < near < far perspective, 0 ≤ near < far orthographic) produces a warning and is ignored. A value left unset is always valid.
• test renderer: reports the far limit instead of a hard-coded 1.0.

ADR 0011 records the design choices: treating far as a plane rather than a sphere, and FLT_MAX rather than +inf when far is unset. The render-pipeline glossary gains Miss, Far Limit and Miss Depth.

The GL device is out of scope: its depth channel is window-space [0,1].

Testing

• TestCameraNearFar runs against the default, debug, fast, interactive and quality renderers. It covers:
  • no clipping when far is unset (misses report FLT_MAX);
  • perspective far (far sphere clipped, misses report far / cosθ) and perspective near;
  • orthographic far, and orthographic near = 0;
  • cutting-plane misses;
  • ray-buffer ray.tmax, and the invalid ray buffer;
  • invalid near/far pairs (warning, no clipping).
• All 54 RTX API test binaries pass locally.