pub fn burn_rms_norm(x: Tensor<2>, w: Tensor<1>, eps: f32) -> Tensor<2>
RMSNorm over the last dim, with weight.