Preserve unlocatable extracts as instance excerpts in nuVal and nuVnu (fixes #38)

[wittjeff]

Fixes #38.

Problem

nuVal and nuVnu use a message's extract only to obtain a data-xpath attribute for the element catalog, then discard it. For messages whose extract is not annotated HTML — notably CSS: Parse Error., whose extract is the erroneous CSS itself — getAttributeXPath falls back to /html, so the standard instance is anchored to the page root and the only informative content of the violation is lost. A report consumer sees CSS: Parse Error. with no indication of what failed to parse or where.

Fix

A new shared helper, getExtractExcerpt in procs/nu.js, returns the extract as an excerpt for the standard instance only when the extract contains no data-xpath attribute — exactly the complement of the case the catalog already handles. So:

• DOM-anchored messages are unchanged: they keep their lean {ruleID, what, ordinalSeverity, count, catalogIndex} shape, with element details normalized in the catalog.
• Unlocatable messages (CSS parse errors and other non-DOM findings) now carry their extract inline as excerpt, truncated at 400 characters (CSS extracts can be whole minified stylesheets).

Both tests/nuVal.js and tests/nuVnu.js use the helper, keeping the two engines' standardization identical.

Verification

• node --check passes on all three files.
• Unit assertions on the helper: CSS extract returned verbatim; data-xpath-annotated extract returns null; missing/empty extract returns null without throwing (e.g. the Nu checker's Too many messages. fatal, which has no extract); >400-char extract truncated with an ellipsis.
• End-to-end run of the nuVal reporter with a stubbed validator response containing one CSS: Parse Error. and one ordinary HTML message:

[
  {
    "ruleID": "CSS: Parse Error.",
    "what": "CSS: Parse Error.",
    "ordinalSeverity": 3,
    "count": 1,
    "catalogIndex": "0",
    "excerpt": ".hero { colr: #fff, }"
  },
  {
    "ruleID": "Bad value  for attribute role on element div.",
    "what": "Bad value  for attribute role on element div.",
    "ordinalSeverity": 3,
    "count": 1,
    "catalogIndex": "1"
  }
]

If you would rather surface the extract on all nuVal/nuVnu instances (not only unlocatable ones), I'm happy to adjust — this version keeps report size flat for the common DOM case.

🤖 Generated with Claude Code