radeonsi: disable compute blit when use aco
ACO generated code does not work when fp16_rtz==true, like piglit test:
spec@arb_copy_image@arb_copy_image-simple --rb-to-rb
I've found one problem that ACO does not support vector nir_f2f16_rtz, like 2x16 ssa1 = nir_f2f16_rtz ssa0.xy. But after fixing this problem the test result only gets improved instead of passing. Maybe there are some image coherency problem as found by piglit test:
spec@arb_shader_image_load_store@coherency
So disable this before we completely fix all the problem.