fix(converter): escape literal fence-like prose so it cannot capture code blocks

[pchuri]

Problem

A prose line that starts with three or more backticks is indistinguishable from a fence the converter emitted itself. splitOnFences pairs it with the next real fence, so the real code body falls outside the fence and goes through whitespace cleanup. The code is changed silently.

htmlToMarkdown('<ul><li>a<ul><li>b</li></ul>```</li></ul><pre><code>  indented\n    code   here</code></pre>')
// the <pre> body loses its indentation

This affects both converters. Besides html → markdown (convert --input-format html, reported in #254), storage → markdown is affected too, which is the read path: a <p>```</p> followed by a code macro loses the indentation of the code body.

Changes

• New shared helper escapeFenceLikeText() in lib/markdown-cleanup.js. It backslash-escapes every backtick of a line-leading run of 3+ backticks in prose (``` becomes \`\`\`). Escaping only the first would leave the rest free to pair with later backticks into a code span on write-back. NBSP (literal or as an entity) counts as indentation before the run.
• storage → markdown: applied in StorageWalker.renderText(), except inside code spans and link labels (which already escape backticks).
• html → markdown: applied before conversion to the prose between tags. <pre> and <code> regions and tag attributes are left alone. Backtick entities (`, `) in prose are decoded first, because the entity pass runs after the fence-aware cleanup and would otherwise bring back a literal fence line. Only < followed by a letter, / or ! is treated as a tag (so prose like a < b is not skipped), and the <pre>/<code> region patterns are used only when a closing tag exists, which avoids a quadratic rescan on unclosed <pre>.
• Code bodies, code spans, attribute values and mid-line backtick runs are unchanged.

Output change

A literal fence-like line in prose is now written as \`\`\` instead of ```. The old output turned into a real code block when written back to Confluence; the escaped form round-trips as prose. Five existing expectations that pinned the unescaped output (#243, #244) were updated to the escaped form. One of them has loose text directly after a callout paragraph with no blank line, which Markdown treats as a continuation of that paragraph, so its write-back assertion now expects that.

Testing

• Unit tests for the helper, including that an escaped line cannot open a fence.
• html → markdown and storage → markdown regression tests: the repro from Literal ``` after a nested list corrupts a later code block in html → markdown #254, a top-level paragraph, <br> before the literal, backtick entities and CDATA, a callout, and round-trip. Also checks that code bodies, code spans, mid-line runs and attribute values are untouched.
• npm test (40 suites, 1578 tests) and eslint pass.

Fixes #254

[github-actions]

🎉 This PR is included in version 2.27.1 🎉

The release is available on:

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

Your semantic-release bot 📦🚀