Try to set ray types correctly for rays -- a little naive now, only
using "camera", "shadow", and "diffuse" for all rays (leave
distinguishing between reflection, refraction, diffuse, glossy for
another day, and currently testrender doesn't support subsurface or
displacement, but maybe someday).

Also, for testshade, improve efficency -- don't decode raytype on
every point.  For each individual shade, testshade was converting the
name of the raytype from a string to a ustring, then decoding to a
bitfield. Pull it all out of the loops by computing the bitfield once.

Adjust glass shaders to not change behavior based on raytype:
The code path that was supposed to suppress noisy caustic rays never
worked properly before because we never set the ray type in
testrender. Now that testrender sets it -- but does not distinguish
between the different types of secondary rays -- the code path is
active, but incorrect, and causes tests to fail. Just remove the
offending clause for now and return to the old behavior of the shader
(no attempt to eliminate caustic rays). Later, we should revamp the
glass shader entirely by switching to the newer MaterialX-inspired
closures.

Note: when you specify a resolution in the Background command in our 
scene xml, it will  run the shader out of any ray context at all (therefore
having no raytype) in order to populate an environment map that will be
used as needed. If you omit the resolution, no importance sampling is
used at all, so it's much noisier. I think the right approach is to do both --
populate the map for importance sampling, but run the shader on the rays
themselves to get higher quality visible results.

Signed-off-by: Larry Gritz <lg@larrygritz.com>