Wire fetch_corpus.py into CI so conformance tests run for real, not skip

[Nathan-D-R]

Every reader's conformance test suite skips silently when its local corpus directory is absent (the corpus lives out of tree / gitignored, by design - not checked into any repo). Confirmed currently skipping on a plain checkout across the whole vendor stack:

• OpenTFRaw: crates/opentfraw/tests/conformance.rs - 'skipping: no Thermo fixture available'
• OpenTimsTDF: crates/opentimstdf/tests/conformance.rs - 'skipping: no Bruker TDF cache present'
• OpenARaw: crates/openaraw/tests/conformance.rs - 'skip: corpus not present'
• OpenSXRaw: crates/opensxraw/tests/conformance.rs - 'skip: corpus not present'
• OpenWRaw: crates/openwraw/tests/conformance.rs - 'skipping: no Waters bundle available'
• OpenMassSpec itself: crates/openmassspec-io/tests/vendor2mzml.rs, crates/openmassspec-io-cli/tests/cli_centroid.rs - 'skipping: corpus not present'

None of these run for real in CI, so no decode path is exercised automatically anywhere in the suite - this is the biggest confidence gap across the readers (a regression in any decoder would only surface on someone's local machine with the corpus checked out).

The fix mechanism already exists: scripts/fetch_corpus.py in this repo is already 'a generalized port of OpenTFRaw's per-repo fetcher' - reads a vendor-agnostic sources.json, resolves PRIDE/FTP URLs, downloads into a local corpus dir, writes a manifest. It's just not wired into any CI workflow.

Proposed scope: check in a small curated fixture corpus (or a CI step that runs fetch_corpus.py against a minimal sources.json - one or two small files per vendor is enough to exercise the decode path) for each affected repo, and add the corpus-fetch step to each repo's ci.yml ahead of cargo test. Suite-wide effort - track per-repo follow-up issues from here once the approach is validated on one repo (OpenTFRaw is probably the best first target, since it already has a partial precedent via its 'validate-mzml' CI job that downloads a single file).

[Nathan-D-R]

Piloted this on OpenTFRaw: Sigilweaver/OpenTFRaw#31

Summary of what landed:

• crates/opentfraw/tests/conformance.rs was always silently skipping in CI (fixture only ever lived in a gitignored, out-of-tree corpus dir). The build job now downloads the same small PRIDE PXD054004 LTQ FT RAW file validate-mzml already uses, into repo-root corpus/ (already gitignored per CORPUS.md), ahead of cargo test, on all three OS matrix legs.
• Chose a minimal, direct curl download over invoking scripts/fetch_corpus.py/the shared fetcher wholesale for this one-file pilot - that machinery resolves URLs through the PRIDE REST API and is built for the full multi-GB, multi-instrument corpus, which is more than a single fixture file needs. The downloaded filename matches the naming convention the shared fetcher would produce, so this stays consistent with a real fetch_corpus.py run if that's what gets adopted for the wider rollout.
• fixture()'s candidate list now checks this repo-root path first, keeping the old sibling-checkout candidates for local dev.
• Verified locally that conformance.rs actually decodes and passes conformance (1057 spectra) with the fixture present, and confirmed in the merged CI logs that the "skipping: no Thermo fixture available" message no longer appears - it's running for real. All three build matrix legs (ubuntu/macos/windows) are green with the curl-based fixture step.

This only covers OpenTFRaw. Follow-up issues should be filed against OpenTimsTDF, OpenARaw, OpenSXRaw, OpenWRaw, and OpenMassSpec itself once this pilot is confirmed working - I haven't filed those yet, leaving that for a deliberate decision on whether to reuse this same minimal-curl pattern or actually wire in scripts/fetch_corpus.py for repos where the corpus fixture needs are less trivially single-file.