chore: bump the engine past the language attribute

[dereuromark]

The pin sat at carve-rs 6f1d9bf, four commits before the {:TAG} language attribute shipped.

[x]{:fr}

<p>[x]{:fr}</p>                      <!-- this package -->
<p><span lang="fr">x</span></p>      <!-- carve-js, carve-php, carve-rs, carve-py, carve-go -->

Same document, two answers, decided by which binding the reader installed. {:fr} is exact sugar for {lang=fr}, {:} is the explicit "language unknown" form desugaring to lang="", and a malformed tag leaves the whole block literal instead of half-parsing it.

Measured in carve-rs 9130a7c - the revision this moves to - against carve-php 520ea41 and carve-js a59df0fa: byte-identical on the sugar, the empty form, the collision with an authored lang=, the padding rule ({: fr} is the empty attribute plus a boolean) and the malformed cases.

The bump also brings the Markdown importer, which is where the new pulldown-cmark, memchr and unicase entries in the lock come from.

Companion bumps: markup-carve/carve-py#38 and markup-carve/carve-go#38 (both merged), and markup-carve/carve-wasm for the same window.

[dereuromark]

Superseded by #34, which is merged.

This branch pins carve-rs 9130a7c7. main now pins 98de7874, and 9130a7c7 is an ancestor of it, 44 commits behind. Merging this as it stands would move the engine pin backwards, taking composite figures (PART 9 section 4c) and the cell-attribute order back out, and the corpus job would go from 0 diverging documents to some number greater than zero.

The language-attribute change this branch was opened for is included in 98de7874, so nothing here is lost by closing it. Measured on main after #34: 1025/1025 corpus documents byte-identical through the built artifact.

[dereuromark]

This would regress the engine pin. Merging it as it stands removes composite figures from carve-wasm.

It bumps to 9130a7c7. Since it was opened, markup-carve/carve-wasm#34 moved the pin to 98de7874, and 9130a7c7 is an ANCESTOR of that by 44 commits - so this is a move backwards, not forwards. carve-rs gained PART 9 §4c in markup-carve/carve-rs#986 (49c513d6) and the cell-attribute rule in #996, both of which land between the two revisions.

Measured: figure_group appears in eleven carve-rs source files at 98de7874 and in zero at 9130a7c7. carve-wasm's corpus went from 94 divergences to 0 on the current pin; taking it back to 9130a7c7 reintroduces the eleven 318-composite-figures documents and the six 319-cell-attributes-bind-after-the-kind-and-alignment-markers documents as failures.

The language attribute this PR is titled for is already included at the newer pin, so nothing here is lost by closing it.