i965/nir: Do lower_io late for fragment shaders

i965/nir: Do lower_io late for fragment shaders

The Vulkan driver wants to be able to delete fragment outputs that are
beyond key.nr_color_regions; this is a lot easier if we lower outputs at
specialization time rather than link time.

(Rationale added to commit message by Ken)

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>