Provides an implementation of `find_objects` that determines bounding
boxes for each label in the image. Currently requires the user to
specify the number of labels they would like to inspect. Raises a
`NotImplementedError` if the user wishes to collect all bounding boxes
for labels. Works by selecting the 1-D positions that correspond to the
label while ignoring all other points. Assumes that these positions
along with an intermediate array of the same size comfortably fit in
memory.

Within a utility function, determines whether any positions were found
for the corresponding label. If not, simply returns `None`. If positions
were found, it manually unravels the positions and finds the maximum and
minimum positions along each dimension. These are stored into `slice`s,
which are packed into a `tuple` and returned. Makes sure to use in-place
NumPy operations to avoid using additional memory.

Try testing `find_objects` with some input data that includes labels
within a chunk, labels that span across chunks, and labels that are not
present. Compare the results of the dask-image function to that of the
SciPy function to ensure they match.

…jects feature branch