bug: `np.ones()` not visible on vispy

[tlambert03]

this is a tricky edge case to be sure, but currently if you show an array of all ones:

import numpy as np
from ndv.util import imshow

data = np.ones((10, 10), dtype=np.uint8)
viewer = imshow(data)

you see nothing at all (at least on a black background). the auto-clims make it so that (1, 1) appears black. This only applies to vispy, pygfx shows it as white. but that would be similarly problematic on a white background canvas.

probably no universal solution (options include using a gradient background, and just assuming dark background) but we should be consistent at least

[marktsuchida]

In MMStudio I think I constrained the clims to always have at least a range of 2 (i.e., the UI enforces min < max), although there might be some regressions when autoscale is used. And when the whole image is the same gray level, autoscale could be special-cased so that it results in the interval ±1 of that value (so that the rendered image is 50% gray; but use [0, 1] or [254, 255] if the whole image is zero or 255 so that it appears black or white). Or the clims could still be allowed to shrink to min == max but the color mapping could have that kind of behavior. Not sure if it's worth the extra complexity/behavioral deviation, though.

[tlambert03]

yeah I did have a local branch that just enforced that max - min was always non-zero. it enforced a range of 1, not 2. but then I gut a little hung up on the magic choice of what exactly to do:

if min == max:
    min-= 1
    # or 
    min -= 0.5
    max += 0.5
    # or something else that was even more aware of exactly what value it was

[marktsuchida]

range of 1, not 2

I think I meant the same (a range covering 2 integer levels).