Rewrite codex() and add phenocycler() for Akoya CODEX / PhenoCycler-Fusion data

[timtreis]

Rewrite codex() and add phenocycler() for Akoya data

codex() only read a 2019 CODEX Processor layout. It read files relative to the cwd, read only the first region, lost the marker names, and could not read any QPTIFF. Current Akoya instruments (PhenoCycler-Fusion) write only .qptiff.

Changes

• New phenocycler() reads PhenoCycler-Fusion 1.x/2.x .qptiff / .er.qptiff scans, including Opal mode. It also reads other PerkinElmer/Akoya QPTIFFs: Vectra Polaris, PhenoImager HT (incl. unmixed uint16), brightfield RGB and inForm component data. Optional .xpd / *channels.csv sidecars go into the metadata, and masks are read via labels=.
• codex() is rewritten for CODEX Processor per-marker TIFFs, Uploader/Processor hyperstacks, Instrument Manager .qptiff scans and the Processor's FCS/CSV cell tables. It also reads masks via labels= or <region>_mask.tif.
• Both readers share one QPTIFF core:
  • Pages are classified by the XML <ImageType>.
  • Level 0 is read lazily through tifffile.aszarr and the pyramid is recomputed.
  • Channel names are de-duplicated.
  • Metadata goes into sdata.attrs["akoya"], both structured and raw.
  • The lazy arrays can be pickled.
• Each region gets its own coordinate system unless the regions share a slide, in which case they are placed at their stage positions in "global".
• The CLI extends codex and adds a phenocycler command.

Breaking (codex())

• Element names change to {region}_image, {region}_cells, {region}_labels and table; they were images and {region}.
• All regions are read; regions=[...] reads a subset.
• var_names are marker names, not cycCCC_chHHH.
• New parameters are keyword-only, and imread_kwargs is ignored with a warning.

Validation

• Synthetic (CI): a QPTIFF, FCS and Processor-tree fixture writer feeds about 110 tests.
• Real public data (tests/test_akoya_real_data.py): it runs with AKOYA_TEST_DATA set and is skipped otherwise. All 10 tests pass on an HPC node. Full-resolution tiles equal tifffile's page decoding, and the same QPTIFF read through codex() and phenocycler() gives the same result.

Data	Generation
Figshare+ 25123538 AML2_183	Fusion 1.x
S-BIAD3467 Tonsil-ctrl	Fusion 2.x, Opal mode
Zenodo 22107401 Arlene_Tonsil_9x9, Panel33C (unmixed)	CIM 1.30; PhenoImager HT
S-BIAD859 reg001_montage.tif	CODEX Uploader hyperstack
Zenodo 7742474 per-marker TIFFs	CODEX Processor 1.7
OME Vectra samples H&E, inForm component data	Vectra Polaris, inForm
phenomenalist tonsil CSV	CODEX Processor table

A one-off check of 23 public datasets found no reader issue. It covered element validation, extents, round trips through sdata.write/read_zarr, spatialdata-plot renders and cell alignment.

Not covered

• HuBMAP layouts (Globus is blocked on our cluster).
• No real CODEX .fcs, .er.qptiff or .xpd files were available; those paths are tested only on synthetic data.