Simplified OptiX intersection programs (#1563)

The OptiX intersection programs for various shapes contained an
unnecessary world <-> object space transformation. This step was not
only unnecessary, but it generates a shockingly large amount of code
related to OptiX motion transforms and all sorts of features that we are
to using at all. All of this logic consumes registers, which has a
negative impact on occupancy of generated kernels.

This commit incorporates a small change to the `get_ray()` function and
adapts the other intersection programs, which brings `optix_rt.ptx` to
14% of its current size.

* SDFGrid: Consider ray's mint (#1564)

* updated PTX code

---------

Co-authored-by: Lovro Nuic <12744626+lnuic@users.noreply.github.com>

Author: wjakob

include/mitsuba/render/optix/ray.cuh

@@ -9,6 +9,7 @@ namespace optix {
 struct Ray3f {
     Vector3f o;         /// Ray origin
     Vector3f d;         /// Ray direction
+    float mint;         /// Minimum position on the ray segment
     float maxt;         /// Maximum position on the ray segment
     float time;         /// Time value associated with this ray
 
@@ -25,13 +26,10 @@ struct Ray3f {
 /// Returns the current ray in instance space (based on the current instance transformation)
 DEVICE Ray3f get_ray() {
     Ray3f ray;
-    ray.o = make_vector3f(optixTransformPointFromWorldToObjectSpace(optixGetWorldRayOrigin()));
-    ray.d = make_vector3f(optixTransformVectorFromWorldToObjectSpace(optixGetWorldRayDirection()));
-
-    float mint = optixGetRayTmin();
-    ray.maxt = optixGetRayTmax() - mint;
-    ray.o += ray.d * mint;
-
+    ray.o = make_vector3f(optixGetObjectRayOrigin());
+    ray.d = make_vector3f(optixGetObjectRayDirection());
+    ray.mint = 0.0f; // Rays cast by Mitsuba always have mint==0.
+    ray.maxt = optixGetRayTmax();
     ray.time = optixGetRayTime();
     return ray;
 }