fix(converter): honor <ol start> in html → markdown

[pchuri]

Problem

storage → markdown has honored <ol start="N"> since #246, but html → markdown (lib/html-to-markdown.js, used by convert --input-format html) always numbered ordered lists from 1, so the two converters disagreed on the same markup.

htmlToMarkdown('<ol start="3"><li>a</li><li>b</li></ol>')
// before: "1. a\n2. b"
// after:  "3. a\n4. b"

Changes

• An opening <ol> now keeps its attributes through tag normalization, and tokenizeLists reads start from it (double-quoted, single-quoted or unquoted). Attributes are parsed one by one and the first start wins, so start=7 inside another attribute's value is not read; the name is matched case-insensitively as in HTML (the walker's XML parser is case-sensitive), and numeric character references in the value are decoded. Only a real <ol followed by whitespace or > keeps its attributes, so <ol-x> and <ol/...> are normalised like any other tag. Keeping the attributes on the tag, instead of collecting them in a separate list, means a list inside a table cell (whose tags are stripped earlier) cannot shift the start of a later list.
• The numbering rules now live in lib/list-start.js and are shared with StorageWalker: resolveListStart moved there unchanged (non-negative integers, at most nine digits including the run of items, otherwise 1), along with the "marker that cannot interrupt a paragraph" pattern. Empty items are not counted, as in the walker.
• A list that does not start at 1 cannot interrupt a paragraph, so it gets a blank line before it, at top level and when nested in a list item. This is the same rule the walker applies.
• The continuation indent was already derived from the marker width, so 9. → 10. indents nested content correctly.

Behavior notes

• Only html input that carries <ol start> produces different output; everything else is unchanged.
• start on <ul> is ignored, and other tags still have their attributes stripped.

Testing

• New <ol start> (#252) tests: the repro, 0, whitespace and leading zeros, quote styles, invalid values falling back to 1, the nine-digit limit, empty items, marker widening past 9, several lists in one document, a list in a table cell before another list, nested and after-text blank-line rules, and an unpaired <ol start>.
• Parity tests run the same markup through htmlToMarkdown and storageToMarkdown and expect identical output.
• npm test (40 suites, 1616 tests) and eslint pass.

Fixes #252

[github-actions]

🎉 This PR is included in version 2.27.2 🎉

The release is available on:

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

Your semantic-release bot 📦🚀