Add default sequential fallback

[NthTensor]

Should be no_std and ideally core-only (no alloc). Falls back to full sequential operation.