fix: offset calculation to handle transformed elements with translateY or translateX in scrollIntoView

[piecyk]

Closes #7716

✅ Pull Request Checklist:

• Included link to corresponding React Spectrum GitHub Issue.
• Added/updated unit tests and storybook for this change (for new code or code which already has tests).
• Filled out test instructions.
• Updated documentation (if it already exists for this component).
• Looked at the Accessibility Practices for this feature - Aria Practices

📝 Test Instructions:

🧢 Your Project:

[hshoja]

Looks great, tested with tanstack virtuliazer and gridlist from react aria 👍🏼

[reidbarber]

Thanks for the PR! Is there an existing story we can test this with? If not, could you create a new one?

[piecyk]

Thanks for the PR! Is there an existing story we can test this with? If not, could you create a new one?

@reidbarber I couldn't find an existing story that demonstrates this issue, so I created a simple example with reproducible steps. Please check it out https://stackblitz.com/edit/vitejs-vite-zmdnq3jq.

[snowystinger]

We're a little nervous about this change. We think it's missing some cases. Looking at some other libraries, there is a fair amount of complexity in determining this. For example https://www.npmjs.com/package/compute-scroll-into-view

We might be able to use this library, though are reluctant to bring in the size increase. So the thing we need to prove is that this change is "enough" for our use cases or decide we're ok with the size increase.

[uhyo]

@snowystinger Hello! Can we get a chance to get this merged?

I just hit the issue when I tried to use DatePicker inside a virtually scrolled container. I really need this fix.

So the thing we need to prove is that this change is "enough" for our use cases

I understand this concern, but I would like to emphasize that the current implementation is actually broken in some sense. The new getBoundingClientRect-based approach might possibly not be perfect, but its intent looks obvious; it seems clearly better than the current naive offsetLeft/offsetTop accumulation approach.

I'd appreciate if you could revisit this change.

@piecyk Thank you for the fix!

[snowystinger]

Thanks, I'll bump with the team to see if we can more concretely define what "enough" means. If anyone has any ideas on how to prove this, we'd welcome the ideas

[devongovett]

Approving for testing. We'll see if we find any regressions.

[yihuiliao]

approving for testing