✅ test: error on missing submodule instead of silent skip

[gaborbernat]

The conformance suites parametrize over glob("tests/html5lib-tests/<subdir>/*.dat"). When that submodule is not checked out the glob matches nothing, so the suites collect zero cases and pytest reports green. Conformance never ran, and nothing said so. A contributor who forgot git submodule update --init, or any CI path that skipped submodule checkout, would ship against a suite that silently tested nothing.

Every submodule-gated module now raises RuntimeError at import when its vendored data is absent, naming the exact command to fix it. The guard sits on the if so a missing submodule fails collection loudly instead of degrading to an empty parametrization. CI checks out submodules (actions/checkout with submodules: true), so the guard never fires there and carries # pragma: no cover on the if line to keep the whole block out of the 100% coverage gate. This covers the three named suites plus the tree-serialize, minify round-trip, markdown, entities, and treebuilder-differential modules, which read the same corpus.

The benchmark corpus under tools/bench-data changes the same way: its silent pytest.skip becomes a hard error. That raise lives inside the test body, which runs only under --codspeed, so a plain pytest run still deselects the benchmarks and is untouched. A --codspeed run without the corpus now fails loudly rather than measuring nothing.

No changelog fragment.

[codspeed-hq]

Merging this PR will not alter performance

✅ 81 untouched benchmarks
⏩ 18 skipped benchmarks¹

---

_{Comparing gaborbernat:test/harden-submodule-gated-tests (ba276fc) with main (dee9eae)}

Footnotes

1. 18 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩