Export annotations in Napari-compatible format

[hannahker]

Based on outputs of #25

Annotations should be associated with a given image slice. So when multiple slices have annotations, the output should be 3-dimensional, rather than flattened into 2D.

Also include in the output:

• What is the label for each class? ({1: "box"}) - these are user-defined names (depends on Add a "custom annotation class" generator #36)
• What colour is assigned to each class? - again, user defined (depends on Add a "custom annotation class" generator #36)
• For the future: brightness and contrast settings for the labelling (depends on Explore how to keep track of brightness and contrast adjustments when user is labelling #38)

[hannahker]

Notes on full vs sparse array? For just the annotations, it is likely that we can just export a sparse array across slices - data isn't required where there is no labelling. Necessary information is about which slice number is labelled. This is also input into the ML model.

Also think about the situation where someone might want to download the full array once it has been annotated - the size might not be that big if a large number of pixels are 0, since this could likely be compressed pretty easily.