The previous implementation would inadvertently reuse the same sample value
for medium distance sampling when rays passed through interfaces between
different scattering media. In turn, this led to significant errors in
rendered images.

See issue #221...

There's no need for the user to see a fatal error, which should be reserved
for an actual internal error rather than a problem with the scene as
specified...

Issue #279.

Add a option to use arbitrary names for grids

Sometimes .nvdb files use arbitrary names for grids. This change adds two options for the user to specify different names for 'density' and 'temperature' grids.

Previously, we'd get NaN values with anisotropic surfaces with zero
roughness in one direction and non-zero roughness in the other.

Fixes issue #283.

Avoid Clang 14 error compiling under Windows

Check to see if a filepath exists before making absolute

Build with Clang 14 under Windows show this error
```
clang-14: error: no such file or directory: '/D'
clang-14: error: no such file or directory: 'PTEX_STATIC'
```
This commit fix this and is also OK for MSVC

Only resolve aperture as a file if not a builtin name

One approach to solving issue #280

Make it consistent with how it's specified with the other media..

And similarly for the corresponding member variable.
"temperaturecutoff" is still accepted for backwards compatibility.

Found by @guoxx, issue #250.

Opportunity noted by @star-hengxing, issue #238.

It's not clear what this would mean and this case isn't handled by
integrators in any case. (cc #271)

Silent most of the MSVC warnings except 4244 and 4146 for non clear reason

For these, if a ray goes through an interface we need to treat the interface
as specular. (This is needed since the shadow ray test used in these
integrators does not handle interfaces but just returns false for shadow
rays that hit them.) Related to #271.

*haveMedia would never be set to true due to a check for a nullptr material
at the start of the function... In turn, this led to the wavefront integrator not
detecting the case of interface materials without participating media, which
still requires special handling for direct lighting to come out right.

This should fix #271.

(Still accept "color" for backwards compatibility.)

Read Optix path from environment.

This way it doesn't has to be set in CMakeCache.txt every time the
repository is checked out.

Add mouse handling.

Noted by @pbrt4bounty in #246.

On Windows, both CPU and GPU can't write to managed memory at the same time.
Previously, it was possible that the asynchronous creation of the sampler, camera,
film, etc., could still be running when OptiX BVH construction started. This was
especially likely with the Halton sampler + random digit permutations, which take
a while to initialize.

Fixes #246.

Cuda toolkit

Add more recent versions, up to 11.7.0, and remove older ones.
11.2.2 is now the oldest version tested.

Allows removing the workaround for NVIDIA's changed signing keys.

taggedptr_test.cpp adds a test type to check if the fix works as expected.