Assuring defined order of consumed random numbers.

We don't want to put zero sized allocations that return nullptr in the
allocations hash table; doing so causes a crash when we prefetch to GPU
memory.

(Not currently used.)

Instead, just push those rays onto the ray queue for the next ray depth.
This gives up to a 5% speed up for scenes with participating media.

No functionality changes.
Cleaned up some of the GPU *WorkItem definitions.

Changes scalar contribution function used in MLT to the luminance of the
sample rather than the raw average of the spectral sample values. Otherwise
just cosmetic changes.

Added io_scene_pbrt to blender export directory readme.

(And a few small cleanups / cosmetic changes.)

Fixes issues with interpolating RGB causing spectra to have different
bounds than if corresponding SampledSpectrums were interpolated.

Issue #82.

Issue #82

When rendering on the CPU but with a build that supports the GPU, there's
no need to copy the RGB->Spectrum tables into newly allocated memory.