Fusion-tools

Features
- Added support for multi-frame RGB images (.ome.tif) (rendering as RGB channels by default)
- Adding support for arbitrary levels of nesting in `PropertyPlotter`

Fix
- Bug fixes for v2.0 layout in components

Features
- Built-in `LocalTileServer` starting in `Visualization` classes
- Incorporating multiple slides and their annotations into the `SlideMap` component
- "Linkage" added for duplicate components in the same main layout. Allows for comparison of multiple slides at the same time and integrating with other components
- More flexibility added to `.utils.shapes.spatially_aggregate` allowing users to specify whether they want to separate out properties aggregated from different structures and whether they would like a summary for each aggregated property or just the mean. (see ./tests/test_spatial_aggregation.py)

Fix:
- Fixed a bug in Jupyter mode so that the actual port specified in __init__ is the one that is used (previously was hardcoded for 8050).
- Removed "port" argument in `LocalTileServer.start()`, conflicts with "port" argument in __init__

Features:
- Adding `load_visium` function to `.utils.shapes`. This enables loading an *.h5ad* formatted *10x Visium* dataset. Barcodes for each spot are added by default but users can also pass a list of *var_names* to include in per-spot properties.
- Users can add aligned object props (from `.utils.shapes.align_object_props`) to a sub-property in the annotation

Fix:
- Debugging some items in `ClassificationDataset` and `SegmentationDataset`

Features:
- Version now starts with 1! 😎

Fix:
- Updating labels returned by `ClassificationDataset` (returning 0.0 for structures that do not include the "label_property")

Features:
- `dataset` module added
- Implements `SegmentationDataset` and `ClassificationDataset` which are PyTorch-formatted Dataset classes which allow for iteration of annotated structures in a set of slides. `SegmentationDataset` returns images and masks while `ClassificationDataset` returns images and labels (stored in GeoJSON properties). PyTorch installation is not required with this implementation, however, to integrate these datasets into a ML-pipeline one of the image/target/label transforms has to convert the data to a Tensor.