[Intel-gfx] Moving intel_decode.c to libdrm

[Eric Anholt]

I was once again embarassed while explaining to either Ken or Paul
about how we handle reusing the intel_decode.c file in two trees.
Here's my attempt at a solution to the problem: Move the code into
libdrm, and try to give it an API that we won't have to continually
rev as we throw the kitchen sink into the intel_decode() function
arguments.

One of the things I'm interested in is doing a version that directly
pokes at BOs instead of just a pointer, which would let us decode
associated blocks as we see the various state pointers to them.
There's also room for some interesting validation in that case.

Further patches (mostly fixing up style) are in my libdrm tree on the
intel-decode branch.  I've tested it with Mesa on gen7 (I have further
code to land to make gen7 decode more reasonable).