intel/nir: Lower memory access bit sizes later

intel/nir: Lower memory access bit sizes later

We're about to do load/store vectorization right before this but we need
that to happen after we've done a round of optimization.  Otherwise,
we'll be getting unoptimized NIR in from ANV and the vectorizer won't be
able to do anything with it.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4367>