Add simple test for center_of_mass

[Mause]

Before you submit a pull request, check that it meets these guidelines:

1. The pull request should include tests.
2. If the pull request adds functionality, the docs should be updated. Put
   your new functionality into a function with a docstring, and add the
   feature to the list in README.rst.
3. The pull request should work for Python 3.5, 3.6, and 3.7. Check
   https://travis-ci.org/dask/dask-image/pull_requests
   and make sure that the tests pass for all supported Python versions.

[GenevieveBuckley]

Ah, I see why you were not sure if the issue contained a legitimate bug.

So, we only see the bug if an array containing integer values is passed to dask_image.ndmeasure.center_of_mass(). The array you make in this test has float values in it, since numpy.random.random() outputs arrays of float values.

For this test to show the bug, it would need to go something like this:

a = np.random.random(shape) * 255 # multiply so values are in 8-bit range 0-255
a = a.astype(np.uint8)  # convert to unsigned 8-bit integer data
d = da.from_array(a, chunks=chunks)

This also means that the actual fix for the issue should also be a single line, converting the array chunks to float type before running the computation in `dask_image.ndmeasure.center_of_mass()

[GenevieveBuckley]

I think the fix should happen in this line of dask-image/dask_image/ndmeasure/__init__.py:

out_dtype = numpy.dtype([("com", image.dtype, (image.ndim,))])

image.dtype needs to be changed to always be float type (instead of just reflecting the type of the input array like it does now). That's probably going to look like this instead:

out_dtype = numpy.dtype([("com", float, (image.ndim,))])

[Mause]

I figured the issue was the data I was testing with, but wasn't sure how to fix it - I'll do that now

[GenevieveBuckley]

Thanks @Mause!