freedreno/a5xx: fdl, CTS fix

In my quest to fix CI times on a5xx, I went looking at a5xx layouts to see if that was the source of our iommu faults that makes us need to run deqp single-threaded. It wasn't, but I brought a5xx closer to a6xx, added some layout tests, REed a couple regs, and fixed a CTS failure.

Based on !5115 (merged) (because testing gles3/31 is too slow otherwise). See also https://github.com/freedreno/envytools/pull/12