gallium: BE formats fix, code generation cleanup
As I continue my efforts (see also !1669 (merged)) to unify Mesa and Gallium format enums, I had to figure out what exactly it is that the gallium formats do on big endian. This involved deleting enough of the generated code that I could see how endianness is currently handled, and I believe I found a significant big-endian bug in the process. Deleting 200KB of code can't hurt our build times, either :)
This series should have no effect on little endian, unless my review of the diffs of the generated code was insufficient.