freedreno: Precompile the likely shader variants.
Inspired by #2782 (closed), start doing precompilation all the time like we did for shader-db, log to ARB_debug_output (aka apitrace replay's stdout) when we have to do a draw-time recompile, and fix the biggest offenders for draw time recompiles in the traces I've tried.
This should reduce stuttering at runtime as apps first use their shaders.