What's the state of this PR?

It fixes the original issue, but does not work well with show h.where(amount: 0.3em): none.
In practice, there's no space after 一、, so show h.where(…): none is usually (if not always) necessary.
Therefore, this PR alone does not fix the real problem, yet introduces a breaking change.

Also the implementation depends on quirks of measure.

I guess there exists a better fix, so I've marked it as a draft.

I think the supposed quirk of measure is rather a quirk of either horizontal spacing or CJ punctuation adjustment. h(0pt) will be discarded during inline layout, while it seems like h(0.1pt) will prevent the adjustment.

From a Western perspective, the patch here seems unproblematic, but if it doesn't really solve the problem for CJ users, that's unfortunate. I can't properly judge it, but from gut feeling perhaps the CJ punctuation adjustment is the root cause and needs change? I'd overall like to see the CJK handling in inline layout reworked as it has various subtle issues and is not that cleanly implemented.

I haven't carefully inspected how the CJ punctuation adjustment is implemented; but based on what DeepWiki told me, I think it looks logical. It's just that the punctuation adjustment doesn't rhyme with higher-level adjustment (e.g., justification).

Higher-level adjustment is irrelevant in this case, because the issue occurs even if justification is disabled and linebreaks are controlled manually.

I believe the root cause lies in this comment:

1. We don't have a locator after the numbering, so we have to resort to measure.
2. The measure infrastructure can't measure things in context. It has to isolate the object before measurement.
3. The space occupied by the content may vary depending on the preceding and following content.