[regression] language-configuration of embedded language not loading

[RedCMD]

some optimization/bug is stopping the language-configuration of embedded languages from loading correctly when part of the embedded language is offscreen during file loading

Create a markdown file and paste in the following code

```latex


()

```

scroll down until you can no longer see ```latex, but still have () visible
reload VSCode (open & close etc)

scroll up and notice how the brackets () are no longer highlighted
typing a random letter near the latex code will start highlighting the brackets
typing a newline wont, nor will deleting any char

❌
✅

seems to also happen when the embedded language is within the visible screen space, but is inside a folded region

this was just showing "brackets", but it also applies to all other settings inside language-configuration.json

seems to be only 90% consistent. prob cause of some caching somewhere
VSCode 1.83.1 is fine ✅
some change in-between
VSCode 1.84.0 has the bug ❌

[RedCMD]

it seems the viewport position only changes the order in which lines update
as you can force the correct (breaking) order without positioning the viewport in any special way

paste this code in a Markdown file

```jsonX
{
	"patterns": [
		{
			"match": "(?x)(?<=((?!)))"
		}
	]
}
```

remove the X from jsonX
notice how none of the brackets have been highlighted 🐛

modify one of the lines inside the codeblock
brackets inside codeblock now highlight ✅

put the X back and then remove it again
brackets still work as they should ✅

it seems the update propagation isn't working correctly the first time a language loads
maybe an await is missing?

it also happens when a single line takes too long to tokenize
which I come across very often

vs

[RedCMD]

it's even worse now in the 1.89 release
typing a single letter now no longer updates the entire line anymore
it seems to update random parts of the line now

only the center 2 )( are highlighted