Fix code editor scrolling experience on track pads

[Kurble]

Fixes #28149

This PR fixes two closely related problems:

• Scrolling slowly in the code editor with a track pad would not do anything;
• Smooth scrolling would animate track pad scrolls, but they are already animated as the pixel deltas are provided on a frame-by-frame basis.

The first problem is caused by the step size that's set on the VScrollBar, any scroll event going through TextEdit::_scroll_up and TextEdit::_scroll_down got rounded down to an integer number of lines. Since scrolling on a track pad happens every frame, the individual deltas are relatively small, and they would be rounded down to zero unless you're scrolling quite fast. This is addressed differently for the smooth scrolling enabled and disabled case:

• When disabled the scrolling deltas are accumulated the same way they were already being accumulated for the smooth scrolling case, so when slow scrolling for long enough the step size threshold is reached;
• When smooth scrolling is enabled the step size is set to 0, allowing all fractional increments.

The second problem becomes quickly apparent when the first problem is fixed: scroll delta's less than 1 line are applied immediately, while scroll delta's larger than 1 line are animated. This feels really snappy as long as the target scroll value is less than 1 line away from the current scroll value, only to suddenly change to a fixed speed when the delta becomes too big. This has been addressed by directly applying any scroll delta coming from InputEventPanGesture events, instead of allowing it to animate.

[Calinou]

Reposting what I discussed on the Godot Contributors Chat:

Rounding scrolling to lines is probably not desired behavior nowadays, even with a mouse wheel (and even without smooth scrolling). For example, VS Code doesn't round to lines when scrolling with the wheel, so I think we can get rid of the behavior entirely 🙂

[Kurble]

Updated to get rid of line snapping.

[Kurble]

Some tests were failing because scrolling to a caret position can now result in fractional scroll values. That actually feels really weird, so I fixed it by rounding the scroll values in set_line_as_last_visible.

The set_line_as_first_visible and set_line_as_center_visible methods are not affected as they deal with integer values.

One test had to be changed since checking for equality of two doubles isn't working there because of a rounding error.

[KeatsPeeks]

Any update on this one ? Without it the editor really feels clunky with a touchpad.

[Kurble]

@KeatsPeeks Currently just waiting for review I guess, PR should be good to go.

[Riteo]

I tested this one on my Wayland branch (#57025) and this works fine even there!

This PR really improves the code editing situation considerably on a track pad and the patch looks concise and clean.

[akien-mga]

Looks fine from the input team's perspective, but we'd like some assessment from those more familiar with TextEdit or the expected behavior of trackpads.

[BoydWiglaf]

@Kurble I created an new pull request #82260 with your changes in it with the updates from the godot master branch. It seems this PR has conflicts. If you have the time to resolve the conflicts I think this PR could get pulled in soon. However, If not I have the new PR with your changes in it. I hope Im not stepping on any toes. I am new to doing these things and really would love to see this update pulled into Godot.
Thanks

[AThousandShips]

@Kurble can you rebase your PR and fix the merge conflicts

[Kurble]

Rebased on master and now the tests even pass without any changes!

[BoydWiglaf]

@Calinou @Paulb23 @bruvzg @godotengine/input @godotengine/gui-nodes @godotengine/tests
What else is needed on this PR?

[Calinou]

I suggest testing the pull request locally and make sure it works on your end. (I don't have a macOS device with a trackpad to test this PR.)

[ikkerens]

Built this from source on my Macbook Pro with M1 Max running MacOS Sonoma 14.0 and found no issues.
Overall this change makes the whole experience a lot more refined and far less sluggish (and actually works when slow-scrolling) than the currently available official release.

I hope this gets merged soon!

[Calinou]

Tested locally (rebased on top of master a574c02) on mouse wheel. It works as expected in the sense that I can't notice any regressions.

Note that this PR doesn't resolve #28385.

[Riteo]

Tested again on top of the Wayland branch (#57025), which implements InputEventPanGesture, and it still works great as last time!

Other than that I'll reiterate that the patch looks really nice and concise, great job!

[akien-mga]

Thanks! And congrats for your first merged Godot contribution 🎉