Summary

HTML examples inside fenced or indented code blocks can disappear from the search index. genIndex() passes their already-lexed text through the Markdown parser again, which treats tags as HTML and removes them.

Keep code-token text as literal content when building the index. Other tokens retain their existing conversion, and search results still use the existing HTML escaping. Regression tests cover a code block immediately after a heading, after paragraph text, at the start of a page without a heading, and an indented code block. They also verify that the HTML example is displayed as text rather than inserted as an element.

To reproduce on a fresh page with the search plugin enabled:

# Code examples

```html
<customwidget />
```

Search for customwidget. Before this change there is no result; after it, the result includes the literal <customwidget /> example.

Related issue, if any:

Reproduced on develop at d823dbe51c1b2dea4c2091e2b952ded0d1e16616; no matching open fix found.

What kind of change does this PR introduce?

• Bugfix

For any code change,

• Related documentation has been updated, if needed (no public API change)
• Related tests have been added or updated

Does this PR introduce a breaking change?

• No

Validation

• Confirmed the new browser regression fails before the fix because the result is missing, then passes after the fix.
• npm run build: passed.
• npm run lint: passed.
• npm run typecheck: passed.
• npm run test:jest -- --runInBand: 128 tests and 46 snapshots passed.
• npm run test:e2e:chromium -- test/e2e/search.test.js --workers=2 --reporter=line: all 23 search-plugin tests passed.

Browser coverage: Chromium via Playwright. Firefox, WebKit, and the full unrelated E2E suite were not run locally.

Developed with assistance from OpenAI Codex; the regression and fix were validated locally.