If the `condition` is shorter than the Dask Array, treat it as if it is
filled with `False` values at the end. In other words, trim the Dask
Array down to the length of the `condition`. This works just as well
when the `condition` is a Dask Array as when it is a NumPy array.

As masked selection is supported by Dask Arrays, NumPy arrays, and other
array types, go ahead and change the `compress` code to use masked
selection in all cases. Should make it easier for other array types to
benefit from this code. Simplifies our code and removes a branch as
well.

If the length of `condition` is shorter, the array will be shortened
with or without this check. Otherwise if the length of `condition` is
the same, then this will be a no-op (and we already test for this). In
the rare event `condition` is longer than our array, this will do
nothing, but we will generate an error later.

Uses the same comprehension trick for truncating the Dask Array above
when slicing with the mask.