improvement: Add support for NIfTI (`.nii` / `.nii.gz`) files across visualization and model workflows

pytc-client currently does not support NIfTI (.nii / .nii.gz) files, which are a standard format in neuroimaging and many biomedical imaging pipelines, most notably being a common format imported into ITK-SNAP.

This issue proposes adding .nii / .nii.gz support across the following components.

Visualization

• Load and render NIfTI volumes directly in the visualization tab
• Correct handling of voxel spacing, affine transforms, and orientation
• Behavior consistent with existing supported volume formats (slicing, zooming, contrast, etc.)

Model Inference

• Accept .nii / .nii.gz files as valid inputs for inference
• Ensure correct preprocessing (axis ordering, spacing, normalization)
• No manual conversion required by the user

Model Training

• Enable training directly from NIfTI volumes
• Support dataset loading, batching, and augmentation
• Maintain compatibility with existing training configurations and workflows

Implementation Notes (suggested)

• Likely leverage nibabel for file I/O and affine handling
• Define a canonical internal volume representation shared across formats
• Clearly document assumptions around orientation (e.g. RAS/LPS), axis order, and units (this might require a subissue?)

Acceptance Criteria

• .nii and .nii.gz files load successfully in visualization
• NIfTI volumes can be used for inference without format-specific workarounds
• Training pipelines accept NIfTI inputs seamlessly
• Documentation added describing supported behavior and limitations

The issue names no files, tests, or entry points. Start by locating the existing volume loaders and format handling for visualization, inference, and training, then trace how spacing, orientation, preprocessing, batching, and augmentation are represented. Done means .nii and .nii.gz work in all three workflows, with documented orientation and limitation behavior.