Fix occasional incorrect fallback rendering

[philippotto]

This PR fixes how the look up buffer for the GPU rendering was written to. The new code only writes a bucket into a slot of the look up texture if there is not already another bucket there with a higher quality. The old code did not check this which is why sometimes data from a coarser mag was rendered even though a finer quality was available.

Todo:

• fix flicker when zooming

URL of deployed dev instance (used for testing):

• https://ensurecorrectfallbackrendering.webknossos.xyz

Steps to test:

• See the reproduction steps in Annotating an anisotropic dataset with volume resolution restrictions behaves incorrectly #6138
• There should be no flickering when zooming in or out (and switching zoomsteps)

Issues:

• fixes Annotating an anisotropic dataset with volume resolution restrictions behaves incorrectly #6138

---

(Please delete unneeded items, merge only when none are left open)

• Updated (unreleased) changelog
• Ready for review

[daniel-wer]

@philippotto I think I found the culprit for the flickering and fixed it :)

I was also able to roughly observe what was happening. See this log from my debugging session (The first number indicates the bucket's zoomstep and the second number posInBuffer. I've focused on a single bucket, the one with posInBuffer 66134):

Zoomed from zoomstep 0 to zoomstep 1 ...
###### Refresh lookup buffer for zoomstep 1
Base bucket fallback written 1 66134
Fallback written 0 66134
###### Refresh lookup buffer for zoomstep 1
Base bucket fallback written 1 66134
Fallback written 0 66134
###### Refresh lookup buffer for zoomstep 1
Base bucket fallback written 1 66134
###### Refresh lookup buffer for zoomstep 1
Base bucket fallback written 1 66134
...

Basically, during the first two refreshes of the lookup buffer, after zooming out from zoomstep 0 to zoomstep 1, the activeBucketToIndexMap still contained buckets in zoomstep 0. These buckets in the finer resolution were treated as fallback buckets (although they are anti-fallback buckets ^^), because isBaseBucket was false for them (zoomStepDifference === 0). As far as I can see, all of the remaining code expected zoomStepDifference to be positive, but it was negative in that case, leading to weird effects which I didn't fully debug.
This bug existed before, but apparently had no effect, because the buckets with zoomstep 0 were never written to the lookup buffer, because the ones with zoomstep 1 were written before (the bug you fixed). Your fix to write buckets to the lookup buffer if they have a better zoomstep brought this bug to light.

[philippotto]

Ah, awesome debugging & find! Focusing on a single bucket is a great idea (which I wish to have had myself back then ^^). I'm very happy that this turned out to be such a plausible bug and not some weird GPU problem :)

https://media.giphy.com/media/3ohzdIuqJoo8QdKlnW/giphy.gif

I'll make the PR merge-ready next.

[daniel-wer]

Thanks for making this PR merge-ready! LGTM 👍