freedreno/ir3: cp postsched (plus misc)

just pulling some stuff out of my wip/vectorish branch which are (mostly) unrelated..

The most interesting part is the last patch which does a post-sched cp pass to eliminate some movs out of "arrays" which are not necessarily safe to eliminate pre-sched. Overall helps shaders with flow control the most (esp. cases where we'd end up with small blocks that read a phi-lowered-to-regs/array and then make a branch decision on it, due to the extra delay slots the unnecessary mov requires). There are some cases where it hurts register pressure, which look like cases where something that splits long live-ranges by inserting a mov would help.