Improve performance of editor tables

[bdach]

• Depends on framework package with Add virtualised list container osu-framework#6312
• Probably get Fix VirtualisedListContainer rows not returning child to pool on disposal osu-framework#6316 in first too
• Fix VirtualisedListContainer rows not returning pooled content in a different way osu-framework#6318
• framework package w/ above

master:

2024-06-26.10-49-33.mp4

this PR:

2024-06-26.10-56-24.mp4

Map used for testing: https://osu.ppy.sh/beatmapsets/1878528#mania/3895383

Aside from improving performance, this also simultaneously fixes utterly stupid issues like this that can happen on master:

which are a consequence of the utterly stupid decision to decouple rows from their backgrounds.

The fun part is all of this is a prelude to doing something about #23147. I just couldn't stand looking at the travesty that is EditorTable (begone) while attempting to figure out how to fix that one.

One nice thing about this diff is that it makes the table headers fixed on screen (i.e. they don't scroll away when scrolling the content). Which I think makes infinitely more sense.

[peppy]

Hmm, this isn't looking great:

[runtime] 2024-06-26 14:11:54 [error]: An unobserved error has occurred.
[runtime] 2024-06-26 14:11:54 [error]: osu.Framework.Graphics.Drawable+InvalidThreadForMutationException: Cannot mutate the InternalChildren on a Loaded Drawable while not on the update thread. Consider using Schedule to schedule the mutation operation.
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Drawable.EnsureMutationAllowed(String action)
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Containers.CompositeDrawable.ClearInternal(Boolean disposeChildren)
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Containers.VirtualisedListContainer`2.ItemRow.Unload()
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Containers.VirtualisedListContainer`2.ItemRow.Dispose(Boolean isDisposing)
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Drawable.Dispose()
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Extensions.IEnumerableExtensions.EnumerableExtensions.ForEach[T](IEnumerable`1 collection, Action`1 action)
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Containers.CompositeDrawable.Dispose(Boolean isDisposing)
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Graphics.Drawable.Dispose()
[runtime] 2024-06-26 14:11:54 [error]: at osu.Framework.Extensions.IEnumerableExtensions.EnumerableExtensions.ForEach[T](IEnumerable`1 collection, Action`1 action)

[peppy]

I don't think this is required anymore.

But in testing that this is correctly handled, I did encounter some weirdness with scrolling into view (which prooobably is not a regression of this PR and can be ignored if so):

2024-06-26.23.14.36.mp4

[bdach]

Turns out hoping that .ScrollIntoView() was going to just-work for this virtualised list thing was too much. From the behaviour I was seeing here it's pretty clear that it just plain doesn't work on rows that haven't been loaded yet, so I replaced it with a local workaround (9384cbc). I'm not super happy about it but still probably worth it for the performance gain?

[bdach]

Hmm, this isn't looking great:

Interesting that this didn't show up framework-side. Probably because I didn't test deparenting thirty levels up like this case seems to be.

I'm not exactly sure what to do with this right now if I'm not allowed to clear a child from this context. I could return it to pool sure but as long as it is in InternalChildren the row wrapper will still try to dispose it.

Will have a think tomorrow.

I did encounter some weirdness with scrolling into view

I thought I saw some at an earlier point of working on this but then couldn't see it anymore. Will check again tomorrow.