Disable network tests whose endpoints or data are gone; fix is_up

[ericprud]

CI runs with SKIP_EXTERNAL_URLS=true, so these only fail locally, but a full local run had 5 failures. For each I looked for an alternate source before disabling it.

Test	What's wrong	Alternatives tried
test_sparql_options::test_named_graph_types	graphdb.dumontierlab.com/repositories/ncats-red-kg returns 502	none found; the expected output is specific to that knowledge graph
test_sparql_options::test_infer_setting	graph.fhircat.org no longer has the fhirontology repository	its 4 remaining repositories (test, test2, example2, ontest) have no Patient/pat4
test_fhir::test_observation_online	hl7.org/fhir/observation-example-haplotype2.ttl is 404	R4 has the example, but R4's observation.shex and full fhir.shex reference undefined fhirpath shapes; R4B and STU3 schemas don't parse. The offline test_observation still covers this case.
test_reactome, test_diseases	Live Wikidata items no longer conform to the 2018/2019 schemas	Validated the 30 committed disease snapshots offline: shex.js and PyShEx agree (6 pass, 24 fail), so this is data drift, not a PyShEx regression

All five are now pytest.mark.skip with a reason.

is_up fix: it treated any HTTP response, including 502 and 404, as "up", which is why the dead GraphDB endpoints weren't auto-skipped. It now treats 5xx and 404 as down. is_down_reason() also always named the last pre-cached server rather than its argument.

Snapshot restore: #113 accidentally included 8 rewritten files under tests/test_issues/data/wikidata/disease/. A local run with network tests on overwrote them (the Wikidata tests save fetched graphs over the committed snapshots), and git commit -a picked them up. This restores them; nothing reads them.

Testing: full suite with SKIP_EXTERNAL_URLS=false: 451 passed, 70 skipped, 0 failed.

🤖 Generated with Claude Code