Implement TextBlock::recreateTextLine and basic multi line support

[evilpie]

Improves #10173

This implements pretty basic multi-line support for FTE's TextBlock. Most of the code was written by Claude using the Weed to Wonder as a test case.

It works by first applying the current format and then measuring the number of characters that fit into the first line. The created TextLine is then limited to the number of chracters that fit. Based on the textLineCreationResult being "success" (line created) or "completed" (all remaining text fit into the current line, no more lines needed) the caller (flashx) decides whether to create new lines or not.

recreateTextLine allows re-layouting a specific line after changing its width. Which means it might fit more or less text than before.

[Lord-McSweeney]

Can start > full_text.len() ever be true?

[evilpie]

So you mean we would usually assume that we either previously consumed less than or at most equal to the length of the full_text? That makes sense. But currently you can still pass random TextLines into that function, which means that property doesn't hold. I imagine when we add more checks we could maybe soften this to ==.

[Lord-McSweeney]

#15159 seems to hang with this PR

[danielhjacobs]

Temporarily switched https://new.weedtowonder.org/ to be built from this branch, can confirm significant improvements.
weedtowonder_current.zip
weedtowonder_PR.zip

[evilpie]

#15159 seems to hang with this PR

Due to the multiple lines we are now executing other code and hanging in this code: https://github.com/apache/flex-sdk/blob/7c47bb9b10c7ca3643079be21597cde022f1bfe1/frameworks/projects/spark/src/spark/components/Label.as#L1544-L1565. Basically getAtomIndexAtCharIndex and getAtomTextBlockEndIndex always returning zero means it just got stuck inside the loop.

[danielhjacobs]

For the failing test, left is Linux with this PR, middle is Adobe Flash Player, and right is the expected image in the repo right now. I think it's fair to say it's not really a regression. It seems like there's a preference for using Flash output or if the Flash output is too different marking the test a known failure and still including the Flash output in the repo, so it may be best to change the test accordingly.

[danielhjacobs]

For completeness, here are the pages above (screenshots with current Ruffle and this PR) taken from Flash's output.
weedtowonder_Flash.zip

[kjarosh]

For the failing test, left is Linux with this PR, middle is Adobe Flash Player, and right is the expected image in the repo right now. I think it's fair to say it's not really a regression. It seems like there's a preference for using Flash output or if the Flash output is too different marking the test a known failure and still including the Flash output in the repo, so it may be best to change the test accordingly.

Maybe I'm missing something but it looks like a significant regression? The text looks cut off at the descender, which also means the line would disappear when we implement FP's text culling logic.