Optimization for Neural Operators can Benefit from Width