feat(nd2): extract instrument-metadata block (objective + acquisition + channels) into audit + OME-XML — replaces #55

[ferrinm]

What to build

Broader-scope replacement for #55. Populate the full instrument-metadata surface on the ND2 reader path so ND2 conversions carry the same audit / OME-XML richness as LIF does after #52 + #62.

Scope

Extract all of the following per scene (or per file where the ADR places them) from an ND2 file — src/zarrmony/readers/nd2.py and the bioio-nd2 surface it wraps:

1. Objective (originally feat(nd2): extract objective lens metadata into audit + OME-XML #55): nominal_magnification, numerical_aperture, immersion, model, working_distance_um. Source: Nikon NIS-Elements structured experiment metadata. Populate per_scene[i].objective + OME-XML <Instrument>/<Objective> + per-image <ObjectiveSettings>.
2. Acquisition + instrument (new from feat(lif): extract per-scene acquisition/instrument block (date, microscope, imaging_method) into audit + inspect #62): acquisition_date, microscope_model, microscope_serial, imaging_method. ND2 records these in the file header / experiment metadata surface exposed by the underlying nd2 package.
3. Channels (from feat(audit): store per-scene channel identity (dye/fluor/excitation/emission) across all reader paths #61): project reader.ome_metadata.images[0].pixels.channels into per_scene[i].channels with the ADR-locked shape.

Implementation notes

• bioio-nd2 exposes ND2 metadata via nd2 (the underlying reader) — check reader.metadata / reader.experiment first, then fall back to a raw walk if the parsed surface is incomplete.
• Normalise Nikon immersion strings (their conventions differ slightly from OME's Immersion enum) — reuse the _normalise_immersion helper shared with the LIF/CZI paths (introduced by feat(lif): extract objective lens metadata (magnification, NA, immersion) into audit + OME-XML #52).
• Normalise imaging_method to the same token set the LIF path uses (widefield_fluorescence, confocal, spinning_disk_confocal, etc.).
• Reuse audit dict shape and OME-XML population helpers from feat(lif): extract objective lens metadata (magnification, NA, immersion) into audit + OME-XML #52 + feat(lif): extract per-scene acquisition/instrument block (date, microscope, imaging_method) into audit + inspect #62.
• Fail-safe pattern from feat(lif): extract objective lens metadata (magnification, NA, immersion) into audit + OME-XML #52. Inspect() output includes the same blocks per scene.

Acceptance criteria

• per_scene[i].objective, per_scene[i].acquisition, and per_scene[i].channels populated on a representative ND2 file.
• OME-XML sibling contains <Instrument><Objective .../></Instrument> and <ObjectiveSettings ID=.../>.
• Same blocks visible in inspect().scenes[i].
• Missing fields → key omitted; missing block entirely → key absent. No dummy values.
• Test with an ND2 fixture (vendor from bioio-nd2's test data if licensing allows, or a small synthetic one).
• Cross-format consistency check: field names + value types match LIF (feat(lif): extract objective lens metadata (magnification, NA, immersion) into audit + OME-XML #52 + feat(lif): extract per-scene acquisition/instrument block (date, microscope, imaging_method) into audit + inspect #62) exactly.
• CHANGELOG entry under a minor version bump.

Blocked by

• ADR-0007: Aperture BigQuery ↔ zarrmony field-mapping contract + instrument-metadata audit shape #60 (ADR-0007 locks the shape)
• feat(lif): extract per-scene acquisition/instrument block (date, microscope, imaging_method) into audit + inspect #62 (LIF establishes the acquisition/instrument block shape)
• feat(lif): extract objective lens metadata (magnification, NA, immersion) into audit + OME-XML #52 (LIF establishes the objective block + OME-XML injection helpers)

Replaces

• feat(nd2): extract objective lens metadata into audit + OME-XML #55 (objective-only slice; closed in favor of this broader-scope reorg per the ingest-team mapping effort)