Wrapper around skimage.pyramid_gaussian using coarsen

I wrote a wrapper around [skimage.pyramid_gaussian](https://scikit-image.org/docs/stable/auto_examples/transform/plot_pyramid.html) that uses [da.coarsen](https://docs.dask.org/en/stable/generated/dask.array.coarsen.html) and other dask and dask-image functions. You can find it [here](https://github.com/samwelborn/tomopyui/blob/main/tomopyui/backend/util/dask_downsample.py).

This is a fix for people who want to pyramid downsample their time series image data (i.e. you have a 3D array with shape Z, Y, X, and the 3D array will be recursively downsampled to Z, Y/2, X/2 --> Z, Y/4, X/4 --> Z, Y/8, X/8). I realized the reason you probably haven't made a zoom function is because there is some C or C++ code associated with ndi.zoom, I guess.

Right now it is totally unclean/disorganized/highly specific because I copied the code from the functional chain in skimage.pyramid_gaussian and applied it for my specific purpose.

I'm gauging interest to see if this is something you would want to add to dask_image. Some of the features in skimage.pyramid_gaussian won't be possible because it doesn't use ndi.zoom, but I could make it a bit more general and submit a PR. Logic is mostly there, just in need of name changes etc.