Capturing and applying a pattern causes performance loss

Create a syntax highlighter with the following code:

{
	"scopeName": "source.redcmd.syntax.lag",
	"patterns": [
		{
			"match": "a",
			"name": "keyword.control",
			"captures": {
				"0": { "patterns": [] }
			}
		}
	]
}

Create a test file, apply that syntax highlighter and type a few thousand a's
notice the increasing lag and vscode eventually giving up trying to match and just highlighting the rest of the file with the last token

(you may need to increase editor.maxTokenizationLineLength to 10,000 or so to see results)

Now remove "patterns": []
notice all lag disappearing and everything tokenizes correctly

Also notice that "patterns": [] can be empty (which would be a different bug in itself) and still causes lag
The capture group itself does need to capture something (does not happen with a empty group)
and the size of the capture text inside the group does not matter

Without "patterns": []; my machine is able to highlight almost 680,000 a's
While with patterns; it can barely highlight 6,500
Thats almost exactly 100x times faster

Reproduce the issue with the shown grammar, an applied syntax highlighter, and a line containing thousands of a characters, comparing the cases with and without "patterns": []. Trace the tokenization path handling captured text and nested patterns. Done means the grammar tokenizes the full line correctly without the reported performance collapse.