Prevent crash when processing line caches in RichTextLabel

[YuriSizov]

Fixes #78912.
Fixes #78849.
Fixes #78932.
Supersedes #78916 as a solution (though sizing changes may be welcome at some point).

I investigated #78241 which introduced this regression and noticed a small logical difference in what appeared to be a simple act of refactoring. We used the old value of scroll_w, valid only at the time of the function call, while in the process the value of scroll_w was updated. This broke pre-existing logic from before the refactoring resulting in the crash.

I admit, I don't really understand the entirety of what's going on, but this doesn't seem intentional and does seem to be the source of the issue.

---

PS. There is also another difference in that refactoring which I haven't touched because it has nothing to do with the crash. But when we do _process_line_caches we used to call main->first_invalid_line.store and main->first_invalid_font_line.store while iterating. In the factored out code of _update_scroll_exceeds this no longer happens, and only main->first_resized_line.store is called. Might be something to look into.

[bruvzg]

There is also another difference in that refactoring which I haven't touched because it has nothing to do with the crash. But when we do _process_line_caches we used to call main->first_invalid_line.store and main->first_invalid_font_line.store while iterating. In the factored out code of _update_scroll_exceeds this no longer happens, and only main->first_resized_line.store is called. Might be something to look into.

Setting this should not be necessary since thread will never exit while inside _update_scroll_exceeds.

[bruvzg]

Supersedes #78916 as a solution (though sizing changes may be welcome at some point).

Note: Sizing changes are still relevant, but should it be tested for the possible Label sizing regressions first (there were a lost of issues).

[YuriSizov]

Note: Sizing changes are still relevant, but should it be tested for the possible Label sizing regressions first (there were a lost of issues).

Yes, though I wouldn't risk it right now, for 4.1. We need to create a gauntlet of unit tests and make sure these changes are safe first. Then maybe pick them for a patch release.

[akien-mga]

Thanks!