WIP: d3d12: Use lower_ubo_vec4 instead of rolling our own.

The previous build_load_ubo_dxil() didn't handle straddling loads correctly, which nir_lower_ubo_vec4() can do for us based on the align_mul/offset fields that GL fills in. We can also avoid addressing math on indirect loads by letting the GL frontend generate load_ubo_vec4s directly.

Super untested, this is just my hacking at "this is where I think d3d12 should be going with ubo loads based on my understanding of the driver"