Fix TTL/RDF generation, working-with-data.md download links, and publish examples/output

[HendrikBorgelt]

Summary

Four related fixes, all touching the example-data/documentation pipeline:

1. working-with-data.md download buttons were broken on the real site. The YAML/JSON download links used href="assets/examples/..." (no ../), which resolves to /working-with-data/assets/examples/... once mkdocs builds the page — a 404. Fixed to ../assets/examples/..., matching the pattern the doc generator already uses for the sunburst-chart iframes elsewhere.

2. TTL/RDF generation was actually fixable, not a pre-existing linkml_runtime bug. linkml-run-examples --output-formats ttl had been failing partway through the example set, and a CI comment attributed it to "a pre-existing linkml_runtime bug (rdflib_dumper mishandling certain CURIEs), unrelated to schema content." Traced it to two concrete, fixable causes instead:

   • 124 occurrences across 5 example-data files used bare QUDT unit codes (unit: GM, unit: PERCENT, unit: g, …) instead of full QUDT URIs. rdflib_dumper tries to resolve every unit value as a URI; a bare code with no colon hits namespaces.uri_for()'s "no prefix" branch and raises Unknown CURIE prefix: @base instead of being treated as a plain string.
   • The three real-world CatalysisDataset records use their published Handle-System identifier as id (e.g. hdl:21.11165/4cat/2d6m-exeb), reference chemical compounds via CHEBI, and one Characterization uses an eNanoMapper (ENM) term for rdf_type. None of hdl, CHEBI, or ENM were declared in the root schema's prefixes: block — they're only declared in vendored chemdcat-ap submodules, which doesn't propagate to SchemaView.namespaces() when the root schema is loaded directly (matching why CHMO/OBI/NCIT are already redundantly declared at the root too).

   With both fixed, --output-formats ttl now succeeds for all 15 example/test files, so it's requested in the _test-examples recipe going forward.

3. Published examples/output/. Its .gitignore rule was recently removed (only the explanatory comment survived) — publishing the json/yaml/ttl round-trip validation output it now produces so it's visible in the repo, and correcting the CI comment that still claimed it was git-ignored.

4. Added a "Download RDF" button alongside the existing YAML/JSON buttons for each of the 3 real-world datasets on the Working with Data page. Publishing these surfaced 8 more of the same root-schema prefix-propagation gaps (RO, RXNO, qudt, schema, skos, foaf, adms, on top of hdl/CHEBI/ENM above) — without them, rdflib_dumper synthesized placeholder ns1/ns2/... prefixes bound to the literal unexpanded CURIE string instead of a real URI. All fixed the same way; .ttl output is now free of placeholder prefixes.

Test plan

• just test passes (39 pytest + schema generation + example validation)
• linkml-run-examples --output-formats ttl produces all 15 .ttl files with zero errors and zero placeholder namespace prefixes
• Verified the download-link fix and the new RDF button locally via mkdocs serve