Extract rule: template-no-whitespace-within-word

[NullVoxPopuli]

Split from #2371.

[NullVoxPopuli-ai-agent]

Review: template-no-whitespace-within-word

Compared against the original ember-template-lint rule no-whitespace-within-word.

What's done well

• The WHITESPACE_ENTITY_LIST is identical to the original -- all 52 entries match exactly.
• The regex construction logic is faithfully ported. The 5-alternation pattern (whitespace)(char)(whitespace)(char)(whitespace) is preserved, including the comment explaining why 5 alternations avoids false positives.
• Parent node filtering for AttrNode and <style> elements is correctly implemented, matching the original's parents.some(...) checks.
• All original test cases are ported -- both good and bad, including the HTML attribute ignore case (enable-background="a b c d e f g h i j k l m"), the <style> element case, and the  /  entity cases.
• Both gjs and hbs parser modes are tested.
• Error message matches exactly: "Excess whitespace in layout detected."

Observations and potential issues

1. Parent traversal approach differs. The original uses path.parents() (provided by the ember-template-lint traversal API) to walk up the tree. The new rule uses a while (parent) loop on node.parent. This is the correct ESLint equivalent and should work, assuming the Glimmer AST nodes have parent references set (which they should in ember-eslint-parser).

2. Uses context.getSourceCode() (deprecated in ESLint 9+). Same note as PR #2592 -- consider using const sourceCode = context.sourceCode || context.getSourceCode(); for forward compatibility.

3. meta.type is set to 'layout'. Note that ESLint deprecated the 'layout' type in ESLint 9 in favor of other types. This isn't an immediate issue but worth being aware of for future ESLint version compatibility. The original rule doesn't have an ESLint-style meta.type so there's no direct comparison, but 'suggestion' might be more future-proof.

4. The regex is compiled at module level (as a constant WHITESPACE_WITHIN_WORD_REGEX) rather than in the constructor as in the original. This is actually better -- it avoids recompiling the regex on every file. Good optimization.

5. CHARACTER_REGEX only matches ASCII letters [a-zA-Z]. This matches the original, but it means spaced-out Unicode words (e.g., accented characters) would not be caught. This is a pre-existing limitation, not introduced by this port.

Summary

Excellent port. The implementation is arguably slightly improved over the original (module-level regex compilation). All test cases are covered, error messages match, and the parent-node filtering logic is correctly adapted to the ESLint API.

---

Automated review comparing with ember-template-lint source