ac/nir/ngg: fix computing the pervertex LDS size for streamout

A shader output isn't necessarily a vec4 like for clip/cull distances.
This computes the pervertex LDS size based on the XFB info instead.

This fixes a bunch of dEQP-VK.transform_feedback.* failures.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>