(Be consistent with changes in #55.)

Flip DiffuseAreaLight's image uv.y coordinate

(This can in particular happen due to image resampling for MIP maps.)

Forgot to add the divide by the wavelengths' PDF here once that was moved
out of Film.

Illuminant photometric scaling so we're back to 0-1 then.

Issue #53.

This reverts commit a685e020, which seems
to have caused issue #56 (the Ubuntu part at least.)

By flipping the DiffuseAreaLight's uv.y texture coordinate we can
match the same texture lookups as SpectrumImageTexture.

In many cases we ended up using a spectrum with the illuminant baked in
when we didn't actually want it.  This caused a blue color shift when equal
RGB values were given to specify scattering and caused the hair BxDF to
come out too red...

(All of these issues were made obvious with the recent
RGBIlluminantSpectrum renaming fixup.)

Fixes, finally, issue #28.

Issue #28.

RGBSpectrum -> RGBIlluminantSpectrum
RGBReflectanceSpectrum -> RGBSpectrum
SpectrumType::General -> SpectrumType::Illuminant
SpectrumType::Reflectance -> SpectrumType::General

(Infrastructure for issue #28)

Avoid refinement depth calc for straight curves

Avoid division by zero with flat ribbon curves

Changed when rays are normalized after being transformed to medium space
which will cause image differences with media that have a scale transform
applied to them. This fixes the visible volume majorant grid artifacts in
issue #28.

Otherwise small changes:
- The cloud medium's "extent" parameter has been renamed to "frequency".
- A small change to how it sums scales noise will lead to small image
diffs with it.
- Otherwise just renaming things / moving them around.

Wrap nvdb filename parameter with ResolveFilename()

This fixes an issue where pbrt can not find NanoVDB files that specified with a relative path.

An example scenario that this fix resolves.
`pbrt scenes/test_file.pbrt`

test_file.pbrt:
```
 # stuff

 MakeNamedMedium "foggy" "string type" "nanovdb" "string filename" [ "../../vdbs/test.nvdb" ]

 # more stuff
```

Now, a trilinear filter.
https://www.iquilezles.org/www/articles/morecheckerfiltering/morecheckerfiltering.htm

Also, do the 3D case.
Inspiration from https://iquilezles.org/www/articles/checkerfiltering/checkerfiltering.htm

Now they use an RGBSpectrum or RGBReflectanceSpectrum, as appropriate, when
turning RGB into spectral samples in their Evaluate() methods.

When a texture is first created, we don't know how it will be
used. Therefore, we create both forms of it and then choose the appropriate
one based on what the Material requests. This eliminates a bunch of cases
where we hard-coded creating ones for Reflectance since we didn't know any
better.

Improve filter footprint computation when no differentials are available and Camera's approximations of dpdx and dpdy are used.

Remove RGBConstantTexture and RGBReflectanceConstantTexture; they're not giving any perf. benefit at this point.

No functional changes.

If a curve is straight we do not need to refine. This avoids a
fatal Log2Int(0) check.