radeonsi: rewrite how occlusion query precision is determined for performance

radeonsi: rewrite how occlusion query precision is determined for performance

The precision of occlusion queries is determined from active queries.
Then the register programming is determined from the precision and other
states.

This has the effect that we no longer set PERFECT_ZPASS_COUNTS
for PIPE_QUERY_OCCLUSION_PREDICATE in some cases, resulting in higher
performance.

This also disables conservative occlusion queries for gfx11 because it's
not recommended with late Z, but detecting late Z vs early Z would be
more complicated, so just never use it, which results in better performance
with late Z.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24732>