chore(tests): run the W3C Notation3 spec suites in CI

[jeswr]

Runs the W3C Notation3 parser and extended suites (manifest-parser.ttl and manifest-extended.ttl) against N3.js as spec-n3-parser and spec-n3-extended jobs in the CI spec matrix. No project-maintained skip list is applied: this PR remains draft and the jobs deliberately expose unsupported cases. Closes #331.

The suites moved to the Notation3 CG, so their canonical manifests are under https://w3c-cg.github.io/N3/tests/N3Tests/ (the old w3c.github.io/N3/tests/… URLs now return 404). manifest-extended.ttl contains parser syntax tests rather than reasoning or evaluation tests.

Why a local checkout + -m instead of -c: rdf-test-suite constructs all test cases concurrently, producing roughly 1,000 simultaneous GitHub Pages fetches and reliable UND_ERR_CONNECT_TIMEOUT failures. PR #612 makes warm .rdf-test-suite-cache runs reliable but cannot safely populate a cold N3 cache. spec-n3-fetch therefore creates a sparse, blob-filtered checkout of w3c-cg/N3 and the suites use rdf-test-suite mapping from URLs to files, preserving their canonical test IRIs without the request burst.

The fixture checkout is pinned to the N3 revision in spec/cache-key.txt. The weekly refresh workflow checks w3c-cg/N3 alongside the RDF 1.1/1.2 suites and proposes revision changes for review, so test updates do not arrive silently.

Current results at the pinned revision: parser 173/222 pass (16 suite-rejected tests skipped); extended 974/978 pass (107 suite-rejected tests skipped). Two parser-manifest TestN3Eval entries are not constructed because the released rdf-test-suite package cannot load their expected-result files.

Validated dependency stack: parser 223/223 pass (16 skipped); extended 978/978 pass (107 skipped). The regular N3.js suite also passes: 20/20 suites and 6,861/6,861 tests with 100% coverage, plus lint.

Merge dependencies

N3.js parser support:

• feat(parser): add opt-in implicit N3 empty prefix #693 — opt-in implicit empty prefix, enabled by the conformance adapter
• feat(parser): support N3 has, is-of, and inverted predicates #694 — has, is ... of, and <- inverse predicates
• feat(parser): support N3 IRI property lists #695 — IRI property lists
• feat(parser): support N3 expression predicates #696 — expression predicates
• feat(parser): support directives in formulas #697 — directives inside formulas

Upstream test infrastructure and fixtures:

• Fix stale parser evaluation results w3c-cg/N3#236 — fixes stale result IRIs and numeric lexical expectations (Parser evaluation results still contain pre-move w3c.github.io base IRIs w3c-cg/N3#234 and numbers.n3 expected result canonicalizes numeric lexical forms w3c-cg/N3#237)
• Add support for N3 expected results rubensworks/rdf-test-suite.js#143 — loads N3 expected results correctly (Load text/n3 expected results for TestN3Eval cases rubensworks/rdf-test-suite.js#142)
• Release rdf-test-suite with Make addPrefix accept URLs that do not end in # or / #143, then bump the dependency here
• After Fix stale parser evaluation results w3c-cg/N3#236 merges, update spec/cache-key.txt to its merged revision

The undeclared-empty-prefix compatibility question is tracked upstream as w3c-cg/N3#235. The current-major opt-in and next-major default are tracked in #693 and #699 respectively.

cc @jeswr