Compute fused RMS normalization: outputi = (inputi / rms(input)) * weighti
result tensor, or null if this implementation cannot handle the input
null