[data grid] improve scrollbar deadzone with overlay scrollbars

[lauri865]

• I have followed (at least) the PR section of the contributing guide.

On OS X and potentially other platforms that hide scrollbars when not scrolling, virtual scrollbars are effectively overlapping. When scrollbar is scrolled to the end, there's a dead zone where you cannot drag the scroll handle at all until you manually scroll it out of the dead zone, and scrolling works in only in one direction (whichever comes last in the DOM, currently horizontal scrollbar).

This PR adds two improvements:

1. Vertical scrollbar comes after horizontal scrollbar, making it render on top. It's more likely that there's more vertical content than horizontal content, hence the handle is smaller and the issue is more pronounced (it's more likely that the whole scroll handle is hidden in the dead zone).
2. On hovering scroll track, the zIndex is incremented by +1, making the deadzone go away. You still cannot scroll from the deadzone itself in one of the directions (previously vertically, now horizontally), but the it's at least an easy win for now, and if you start hovering/scrolling outside of the dead zone, it works in both directions now.

[lauri865]

The issue:
https://github.com/user-attachments/assets/ca67bd65-99e2-4e56-89cc-58e787cbf165

[mui-bot]

Deploy preview: https://deploy-preview-15961--material-ui-x.netlify.app/

Generated by 🚫 dangerJS against c96cdfe

[romgrk]

Fyi, about the case that makes the CI fail, sometimes those act() warnings come from a different case in the same file than the one marked as failing, you can try to pinpoint which one by using it.only(...). act() is a react/RTL hack so it's finicky to deal with, not sure why this diff triggers that warning.

[lauri865]

All the tests were passing when I first opened this PR. It's very unlikely that such a simple change would break any tests to begin with.

Since then, merged master to the PR twice:

1. First time Argos test failed (scrollbar size significantly different)
2. Second time Argos test passed (no change to this PR), but the above failed (which might not pop up when merging again).

Seems unrelated to the nature of the change, so I don't feel like debugging the test system to be honest.

[lauri865]

Now there are 576 Argos changes 🫠

FWIW, all the unit tests are (still) passing locally, so I'm not sure how I could even debug the CI failure to begin with.

[romgrk]

The argos issue is present in all open PR:s, should be dealt with shortly. And if the other failure is a flaky test we'll ignore it and deal with it independently.