MattesMutualInformation corner case spacing error

### Description
corner case error in ITK's MattesMutualInformation metric when both images are all zeros, then the internal data structure (Image) has the reported invalid size

### Steps to Reproduce

> import SimpleITK as sitk
> import numpy as np
>
> I = np.zeros((100,100,100))
> I = sitk.GetImageFromArray(I)
> I.SetSpacing((1.0,1.0,1.0))
> I = sitk.Cast(I, sitk.sitkFloat32)
>
> J = np.zeros((100,100,100))
> J = sitk.GetImageFromArray(J)
> J.SetSpacing((1.0,1.0,1.0))
> J = sitk.Cast(J, sitk.sitkFloat32)
>
> R = sitk.ImageRegistrationMethod()
> R.SetMetricAsMattesMutualInformation()
> MMI = R.MetricEvaluate(I, J)

### Expected behavior
Evaluation of Mutual Information Metric

### Actual behavior

> ERROR: Image(0x32256f0): A spacing of 0 is not allowed: Spacing is [0, 0]

### Reproducibility

100%

### Versions

### Environment

Ubuntu 18.04, python 3.7, SimpleITK 1.1.0, numpy 1.17.2

### Additional Information

Changing metric e.g. SetMetricAsCorrelation() does not throw error -> therefore must be the MattesMutualInformation metric
Initializing image with np.random.rand(100,100,100) does not throw error -> therefore must be the corner case where inputs are zero