Subroutines breaking capture tokenizing inside of referenced capture group

When trying to call a subroutine on a capture group via \\g<1>.
The call will remove all the previous tokens from capture groups that aren't rechecked in the subroutine.

Create a syntax highlighting extension with this code

{
	"$schema": "https://raw.githubusercontent.com/martinring/tmlanguage/master/tmlanguage.json",
	"name": "Subroutines Syntax",
	"scopeName": "source.redcmd.syntax.subroutines",
	"patterns": [
		{ "include": "#subroutines" }
	],
	"repository": {
		"subroutines": {
			"match": "((a)|(b)|(c)|(d))-\\g<1>",
			"captures": {
				"2": { "name": "strong variable.other.constant" },
				"3": { "name": "strong keyword.control" },
				"4": { "name": "strong support.type" },
				"5": { "name": "strong constant.character.escape" }
			}
		}
	}
}

Expected outcome is that it will highlight all text in the format [abcd]-[abcd]

a-a
a-b
a-c
a-d
b-a
b-b
b-c
b-d
c-a
c-b
c-c
c-d
d-a
d-b
d-c
d-d

Like so:

But instead all tokens connected to capture groups that don't get rematched against (and fail) in the subroutine call get purged.
(capture groups 2 to 5)

Reproduce the issue with the JSON grammar and a-a through d-d examples provided in the report. Start by tracing how the tokenizer handles the \\g<1> subroutine and capture-group tokens; done means the expected capture-group highlighting remains visible for all listed combinations.