Read the JCAMP-DX version as a version, not as the whole record line

[gdevenyi]

Fixes #197.

Spec 2.1 makes $$ a JCAMP-DX comment. _detect_version returned the whole line tail and compared it against a list of literals, so a real spnam1 was rejected:

$ head -2 resources/testdata/pv7/full/20210128_122257_LEGO_PHANTOM_API_TEST_1_1/7/spnam1
##TITLE= /d/exp/stan/nmr/lists/wave/bp
##JCAMP-DX= 5.00 $$ BRUKER JCAMP library (alpha version)

JcampdxVersionError: "5.00 $$ BRUKER JCAMP library (alpha version)" is not a valid JCAMP-DX version

Folder.make_tree catches JcampdxVersionError and drops the child, so through a Folder the file just vanished with no diagnostic. The whitelist already carried two library-identification strings rather than versions — the same problem, patched one spelling at a time.

Change

JCAMPDX.parse_version cuts at $$ and takes the first token. _detect_version, the loaded-parameter path in version, and verify_version all go through it, so SUPPORTED_VERSIONS becomes four version numbers.

After the fix both spnam0 and spnam1 in the PV6 and PV7 studies load and report 5.00; acqp still reports 4.24.

Test

test_the_version_record_is_read_as_a_version_not_as_a_whole_line covers all five spellings observed across the corpus, and test_an_unsupported_version_is_still_rejected keeps the rejection path honest. Both build their file inline — no vendor data.

Suite: 2146 passed, 12 skipped.

[gdevenyi]

Rebased onto #213 (ruff format), so this branch carries that commit until #213 merges — merge #213 first and this diff reduces to its own change. Verified after the rebase: ruff check and ruff format --check clean, full suite green.

Integration check across all eleven conformance PRs merged together: 2164 passed, 12 skipped; corpus load test 3202 → 3207 of 3478 with zero newly broken datasets.