llvmpipe framebuffer fetch support
adds framebuffer fetch support. I haven't added coherent support, it might be easy enough, but not sure it's worth it yet.
I'm also not sure whether the load_unswizzled_block changes needs to be a separate function, and although the bit shift math is correct, it may be overkill, I really had a hard time locking in the correct answer for swizzling the incoming loads into the correct execution order for 8-wide.