ifcb_extract_biovolumes() and ifcb_summarize_biovolumes() gain a custom_cell_counts argument, so use_cell_counts = TRUE (and with it carbon_conversion = "cell") now works with custom_images/custom_classes, which previously could not use chain-count abundance or per-cell carbon at all. The counts are matched to custom_images by position and follow the same semantics as counts read from classification files, including NA marking a ROI without chain-count data, which makes the affected sample’s cell_countsNA rather than a silent partial sum. Invalid input (wrong length, all-NA, counts without custom_images or without use_cell_counts = TRUE) fails with an error saying what to change.
Minor improvements and fixes
.h5 classification files are read several times faster, by skipping a full garbage collection per file.
A damaged or incomplete .h5 classification file now stops with an error naming the file and what is wrong with it, instead of a misleading class_labels_auto does not exist or a raw HDF5 error.
Fixed the scores of single-class .h5 classification files being read transposed.
