Function should_quantize 

pub fn should_quantize(shape: &[usize], dtype: Dtype) -> bool

Whether a tensor should be quantized: rank >= 2 (so biases/norm weights are left alone) and a dtype crate::read::read_f32 can upcast from.