pub fn all_weight_bytes(spec: &ModelSpec) -> Result<Vec<Vec<u8>>, String>Expand description
Read every weight file of a model, in registry order, for native inference.
Multi-file models (e.g. NeXtQSM’s BFR + VJP U-Nets) need all pieces; the order
matches ModelSpec::files, which the per-model glue relies on. Same
local-first/download resolution as primary_weight_bytes.