bidsmri2nidm: field-map capture + BIDS-inheritance sidecar metadata

[dbkeator]

bidsmri2nidm: field-map capture + BIDS-inheritance sidecar metadata

Two related fixes to the legacy (prov-toolbox) bidsmri2nidm converter, rolled
up for the 4.5.5 maintenance release. Fixes #453.

1. Capture BIDS field-map scans (fb4ead7)

• Build the BIDS layout with validate=False so files under fmap/ (which many
  real datasets lay out non-conformantly) are indexed instead of skipped.
• Type field-map acquisitions with a nidm:FieldMap image-usage type
  (BIDS_Constants / Constants additions), including the ABIDE-II case where a
  field map is misplaced inside dwi/ — so it is no longer mislabeled
  DiffusionWeighted.
• Emit the BIDS run index on anatomical scans (parity with the func/dwi branches).

2. BIDS-inheritance sidecar metadata via pybids get_metadata (7007f83, fixes #453)

bidsmri2nidm previously read scan metadata two ways: a per-scan sidecar read,
plus a hardcoded descent into a dataset-root T1w.json (anat) /
task-rest_bold.json (func). That hardcoding only covered T1w anatomicals and
rest-task functionals, and the two reads were applied additively with no
precedence (a scan-level value and a conflicting dataset-root value could both be
emitted).

@seanhatton flagged this in #453; that PR generalized the filename but read the
per-scan sidecar a second time (duplicate) and dropped the dataset-root
inheritance entirely. This fixes the root cause instead:

• New _scan_metadata(bids_layout, directory, file_tpl) →
  bids_layout.get_metadata(file_tpl.path), which implements the BIDS
  inheritance principle (dataset-root + session + scan-level JSON merged, for any
  suffix/task/session), with a direct-sidecar fallback.
• All per-scan sidecar reads route through it; the hardcoded T1w.json /
  task-rest_bold.json root blocks are removed (now redundant).
• Scan-level sidecar values correctly override dataset-root values.

(Also silences a py3.12 f-string E231 false positive on one pre-existing line
via # noqa.)

Testing

Full tests/experiment suite passes: 103 passed, 3 skipped.

---

Thanks @seanhatton for catching the sidecar-metadata issue — this supersedes
#453, which I'll close in favor of this PR.

Note: this is a maintenance patch on the retiring 4.x line; the field-map term
here (nidm:FieldMap) is refined in the forthcoming v5.0.0 LinkML release.