WIP: nir/ntt/virgl: move atomic variable offset into intrinsic value range_base and make use of this in ntt+virgl
The virglrenderer host needs the original offset to pick the right array when addressing atomics, so pass them through.
Fixes: KHR-Single-GL43.arrays_of_arrays_gl.AtomicUsage on virgl
WIP: because the CI expectations need update and it is on top of !19980 (merged)