[Bf-cycles] Volumetric patch update

[storm]

Hi.

I fear that there must be at least one big rewritting from scratch, and
any cvs will be not useful because difference will be high. That code is
more like "see now, i can write bidir render too" than real working
example. I have most features resolved in mind, but many still too
early. One is that use one universal single path weight function, as
Luxrender and other, maybe with many parameters, but current state where
it duplicated in many places is awful. Other is more clean camera and
light evaluators and samplers, it hacked so badly, it is leftover of my
early attempt to get proper camera pdf, spreaded over all kernel_mlt.c .
Logic how to serve perfect mirrors and transparent surfaces is very
complicated too, need to invent something simple and manageable.

In other words, for me it is like tool to do real test "Path tracer vs
Bidirectional tracer", to answer question, is Arnold way is better, or
there is potential to bidirectional and maybe more complicated sampling
technic (some volume as sampled domain?).

I think i can get close to git/svn ready state in 2 months, will see.
Last little success with MIS give me boost in hacking, as many test
scenes start looking better.