pub fn compute_relu(input: &[f32], output: &mut [f32])
Compute contiguous Float32 ReLU with MLAS SIMD.