gallium, st/mesa, draw, util: fix perf defects, support NIR in draw, don't generate TGSI, ST_DEBUG cleanup, driver fixes for a shader CAP

These are various cleanups and improvements:

• The fix for the performance defect in st/mesa is somewhat important.
• UTIL_MAX_CPUS replaced with precomputed align(nr_cpus, 32) to significantly decrease overhead in set_thread_affinity.
• By adding NIR support into the draw module, st/mesa doesn't have to generate both NIR and TGSI anymore. (time to kill TGSI in st/mesa?)
• PIPE_SHADER_CAP_SUPPORTED_IRS reporting is fixed in 7 drivers.
• Some ST_DEBUG options are removed.