Fix _on_view_ndims_requested

[tlambert03]

And in this case, the widget leak is just a failed cleanup after the actual error that is shown a bit higher (https://github.com/pyapp-kit/ndv/actions/runs/15305150376/job/43055857610?pr=202#step:11:393)

Ah, good catch. Yeah, looks like an issue likely arising from the changes added by #196. I'm guessing it's a race condition between a 3D data request and clearing the canvas (i.e. the ImageHandle backed by a vispy image is being used for displaying 3D data when it should have been cleared and replaced by a vispy volume ImageHandle?)

Originally posted by @gselzer in #202 (comment)

@gselzer, I'm not totally following what broke here yet. Is this something you're working on?

[gselzer]

@gselzer, I'm not totally following what broke here yet.

That's just my hunch, based on the stack trace, as to why we get these problems intermittently 🤔

Is this something you're working on?

Not something I'm currently working on, no.

[tlambert03]

no you were definitely right. git bisect puts it solidly at #196. I think it's assuming synchronous behavior and something is out of sync by the time we draw. it's blocking most of our tests now so I might possibly revert #196 if it's too hard to figure out, and we can add back the nice features after we figure out the test failures

also, for me it's not intermittent. I get it every time