freedreno: asynchronously compile initial shader variants

The first 9 patches are various cleanup and prep, which I think makes sense irrespective of asynchronously compiling initial shader variants. The last patch is the actual async support.

Doesn't completely get rid of the janks when asphalt9 compiles new shaders mid-game, possibly they are used too soon after they are created? At least subjectively it feels like it improves game startup time. When combined with threaded-ctx (#4233 (closed)) I think this should be able to hide the recompile latency.