fix(metadata): lenient RDFa extraction for unbound XML prefixes (#4)

[vijaya-boddipudi]

Fixes #4

Summary

PSMetadataExtractorService (RDFa / Semargl parse over published HTML) previously threw SAXParseException: The prefix "gcse" for element "gcse:search" is not bound when published pages included vendor embeds such as Google Custom Search without an xmlns:gcse declaration. That failure aborted the whole page's metadata delivery in PSMetadataDeliveryHandler with ERROR even when file publish itself succeeded.

This change makes extraction lenient:

1. Pre-sanitize HTML before RDFa parse: collect xmlns:* declarations on the document, strip elements and attributes whose prefixes are not declared (deepest-first via Jsoup.unwrap, so children move to the parent before the parent is unwrapped), and rewrite non-XML named entities to numeric character references.
2. Defensive fallback: wrap the RDFa parse in a try/catch that detects SAXParseException / unbound-prefix parse messages and logs WARN with the page path and the offending prefix when extractable, so the rest of the page metadata still flows through.
3. Normal dcterms:* / og:* / perc:* metadata is unchanged because the metadata of interest lives in attribute values, not in unbound element/attribute names.

Acceptance criteria

• HTML with unbound gcse:search (and similar) does not throw out of PSMetadataExtractorService / fail the delivery handler with ERROR for that reason alone.
• WARN identifies page path + problem prefix for support (the pre-sanitize WARN and the catch-around-parse WARN).
• Normal RDFa / dcterms metadata still extracted when present (verified by the new testUnboundPrefixGcseSearch fixture, which also asserts dcterms:source, dcterms:title, dcterms:description, dcterms:abstract).
• Unit test covering gcse (and generic unbound prefix) fixtures.

Files changed

• system/business/src/com/percussion/delivery/metadata/PSMetadataExtractorService.java — pre-sanitize (stripUnboundPrefixedMarkup, rewriteNonXmlNamedEntities) + lenient catch around RDFa parse (isUnboundPrefixParseFailure, extractUnboundPrefix).
• system/Testing/src/com/percussion/delivery/PSMetadataExtractorServiceTests.java — 4 new tests (full fixture, minimal inline HTML, positive / negative helper coverage).
• system/UnitTestResources/com/percussion/delivery/unbound-prefix-gcse.html — new fixture with <gcse:search>, vendor:data-id attribute, and dcterms:* / og:* metadata.
• CHANGELOG.md — new 8.1.8 / GH_POST_PR_COMMIT_RUN_ID entry per AGENTS.md guidelines.
• AGENTS.md — point at main (the active branch), not the non-existent development-8.1.x.

Verification

./mvn-env.sh test -pl system -am -Dtest=PSMetadataExtractorServiceTests -Dsurefire.failIfNoSpecifiedTests=false

Tests run: 9, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 3.949 s
  - testUnboundPrefixGcseSearch                       (new)
  - testUnboundPrefixOnlyDoesNotThrow                 (new)
  - testIsUnboundPrefixParseFailureDetectsGcse        (new)
  - testIsUnboundPrefixParseFailureIgnoresOtherErrors (new)
  - testEntityAndScriptHandling, testbwcorona, testNoAbstract,
    testgetPlainProperty, testgetNamespace             (existing, all pass)

Spotless check on the modified files: clean. Checkstyle: pre-existing config issue on main (JavadocMethod.scope property) — not introduced by this change.

Related

• Sister issue on the main branch: Metadata extraction fails on unbound prefixes (e.g. gcse:search) during DTS publish percussioncms#2355
• Sister fix PR: fix: lenient metadata extraction for unbound XML prefixes (#2355) percussioncms#2375

[natechadwick-intsof]

Summary

This is a solid, JDK 8-compatible backport of the already-merged sister fix (percussioncms#2375). PSMetadataExtractorService now unwraps undeclared XML-style prefixed elements/attributes (the gcse:search case) before XHTML serialization and, if SAX still reports an unbound prefix, swallows that parse failure so DTS metadata delivery no longer ERRORs the whole page. The strip logic matches Jsoup 1.23 (tagName() is the qualified name, e.g. gcse:search), reserved xml/xmlns prefixes are kept, and dcterms/og values live in unprefixed attribute values so they are not stripped. Dominant residual risk is that the new fixture places every asserted RDFa node before the unbound markup, so a broken sanitizer plus the new catch would still go green; the fallback matcher is also a bit broader than the sister PR.

Issue counts by severity

• bugs: 0
• suggestions: 2
• nits: 0