fix(walker): preserve <u>/<sub>/<sup>/<mark> on storage ↔ markdown round-trip

[pchuri]

Summary

Picks option 2 from #155 (selective tag whitelist) for <u>, <sub>, <sup>, <mark>. These four inline tags now survive storage ↔ markdown ↔ storage end-to-end. Closes #155.

What changed

• lib/storage-walker.js — adds the four tags to the existing <details>/<summary> raw-HTML pass-through case, so storage→markdown emits e.g. H<sub>2</sub>O instead of dropping the wrapper to H2O.
• lib/macro-converter.js — wraps markdownToStorage / markdownToNativeStorage with a stash/restore that hides whitelisted tags from MarkdownIt (which is constructed with html: false) and re-injects them after rendering. Code fences and inline code are stashed first, so a literal <u> inside a backtick block stays escaped as text rather than being passed through.
• tests/macro-converter.test.js — 11 new tests covering walker output for each tag, full markdown→storage→markdown round-trip, attribute preservation, the <script> escape (proves the whitelist boundary holds), and the inline-code / fenced-code carve-outs.

Why this approach over the alternatives in the issue

• Option 1 (html: true on MarkdownIt) — too broad. It would let arbitrary user HTML reach Confluence storage XHTML and rely on Confluence's sanitizer as the safety boundary. The four tags above are the only documented information loss; opening the whole HTML grammar to address it is disproportionate.
• Option 3 (status quo) — leaves real information loss for chemistry/math docs (<sub>/<sup>) and any <u>/<mark> use, including content created via the Confluence editor.
• Option 2 (this PR) — fixes the four tags the issue identifies, keeps the safety boundary surgical, and re-uses the STASH_DELIM pattern already in setupConfluenceMarkdownExtensions. No new dependencies.

<details>/<summary> were mentioned parenthetically in the issue but deferred — they're block-level and need different placeholder shaping to avoid <p>-wrap interactions. Out of scope here.

Verification

npx jest          # 612 passing (was 601; +11 new)
npx eslint lib/macro-converter.js lib/storage-walker.js tests/macro-converter.test.js   # clean

Reproduction from the issue, after the fix:

const c = new MacroConverter({ isCloud: true });
c.storageToMarkdown('<p>H<sub>2</sub>O</p>');                  // 'H<sub>2</sub>O'
c.markdownToStorage(c.storageToMarkdown('<p>H<sub>2</sub>O</p>'));  // '<p>H<sub>2</sub>O</p>\n'
c.markdownToStorage('<script>alert(1)</script>');              // '<p>&lt;script&gt;alert(1)&lt;/script&gt;</p>\n'
c.markdownToStorage('`<u>x</u>`');                             // '<p><code>&lt;u&gt;x&lt;/u&gt;</code></p>\n'

Test plan

• New unit tests for each of the four tags (storage→markdown direction)
• Round-trip tests for <sub> and <mark>
• Whitelist boundary test: <script> is still escaped
• Code carve-out tests: <u> inside inline code and inside fenced code is preserved as literal text
• markdownToNativeStorage shares the same passthrough policy
• Full npx jest suite (612 passing, no regressions)

[github-actions]

🎉 This PR is included in version 2.3.1 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀