[DAR-6557][Internal] Use np.uint8 as datatype for output nifti volumes to reduce allocations

[vvihorev]

Problem

Export to nifti volumes has a large memory footprint. We encountered out of memory errors during export.

Solution

Each exported nifti volume is casted to int16 to be written to disk, moreover, each volume we export has just one mask. Thus we can use np.uint8 datatype for volumes during export processing to save space.

Changelog

Use np.uint8 as datatype for output nifti volumes to reduce allocations

[linear]

DAR-6557 Optimize memory footprint for nifti export

[Copilot]

Pull Request Overview

This PR reduces the memory footprint during nifti volume export by switching the internal pixel array data type from the default (implicitly int16) to np.uint8. Key changes include:

• Updating np.zeros calls for the output volumes and multilabel volumes to use dtype=np.uint8.
• Using np.zeros_like calls with dtype=np.uint8 in mask conversion.