pub fn zimage_preload(
model: &Arc<CmfModel>,
geom: &ZGeom,
noise_refiner: &[ZBlockRef<'_>],
layers: &[ZBlockRef<'_>],
context_refiner: &[ZBlockRef<'_>],
) -> boolExpand description
OPTIONAL (B2): build the backend’s weight planes for the per-step blocks
and the context refiner ahead of zimage_prepare, so the caller can
overlap the upload with the (CPU) text encoder. false = not done;
zimage_prepare builds whatever is missing either way.