Dimension order for multichannel vs RGB images

[manzt]

In #1, I noted that we should prefer the last two dimensions of the tiled arrays to be [y,x] as this is most efficient for accessing individual channels.

However, after looking at some whole slide images, I've found that RGB and RGBA images are mostly commonly stored as [x, y, rgba]. This means that for a 2 x 2 image, the data are stored interleaved like:

[r1, g1, b1, a1, r2, g2, b2, a2, r3, g3, b3, a3, r4, g4, b4, a4] 

rather than

[r1, r2, r3, r4, g1, g2, g3, g4, b1, b2, b3, b4, a1, a2, a3, a4]

This makes sense because generally we will view rgba for a particular [x,y] slice together. napari guesses whether an array is RGB by looking at the shape of the array. If the last dimension is 3 or 4, it guesses RGB/RGBA, otherwise the last two dimensions are decided to be [y,x].

We should tile the last two dimensions preferably, but for RGB/RGBA images we should leave this last dimension as channels since this is most natural for WSI. We can use a rgb attribute in the base array to tell create_pyramid.py which dims to downsample and tile.