enable-referable-subfigures`: cross-references to a subfigure's `ref` supplement hardcode "Panel" and omit the parent figure number

Description

When referencing a subfigure (created via subfigure(...) inside enable-referable-subfigures) from a context other than plain in-line body text, specifically, from inside the caption: of a different figure, the rendered reference does not match the normal in-body behavior.

Expected: referencing a subfigure prints something like Figure 6(a), combining the parent figure's number with the subfigure's letter (as it does when referenced from ordinary paragraph text).

Actual: the reference instead prints Panel (a) — a hardcoded "Panel" supplement, with no parent figure number at all.

Minimal reproduction

#import "@preview/patstdlib:0.3.3": enable-referable-subfigures, subfigure
#show: enable-referable-subfigures

#figure(
  grid(
    columns: 2,
    subfigure(image("a.png"), caption: [], label: <fig:main_a>),
    subfigure(image("b.png"), caption: [], label: <fig:main_b>),
  ),
  caption: [Main figure],
) <fig:main>

#figure(
  image("c.png"),
  caption: [
    Compare with the structure shown in @fig:main_a.
  ],
)

In the second figure's caption, @fig:main_a renders as Panel (a) rather than Figure 1(a).

Additional note

The supplement appears to use a non-breaking space between "Panel" and "(a)" rather than a regular space, which also makes it hard to str.replace/regex out as a workaround (had to match \s rather than a literal " ").

Ask

Either fix the reference resolution so it's consistent regardless of the caller's location (matching in-body behavior), or document the limitation and expose a supported way to override the supplement / get the combined "Figure N(letter)" form manually (e.g. a helper function), since #ref(label, supplement: ...) currently appears to be ignored for subfigures.

Start with the minimal reproduction and trace the enable-referable-subfigures, subfigure, and #ref behavior when a subfigure label is referenced inside another figure's caption. Done means the caption reference resolves consistently with body text as Figure N(letter), or the documented supported override/helper produces that form.