Route illumination workers to GPU queue

[arjunrajlab]

Summary

• route both illumination-correction workers to the GPU Celery queue with isGPUWorker=true
• document that the label controls queue placement rather than claiming CUDA acceleration
• update the worker registry GPU column for both images
• add regression coverage so future Dockerfile changes cannot silently route either worker back to the CPU queue

These are long-running image-processing jobs. Keeping them off the CPU queue prevents them from occupying the workers that serve interactive --request interface calls.

TDD

• added GPU-label assertions first and confirmed they failed against the existing isGPUWorker=false labels
• changed both production Dockerfiles and confirmed the routing assertions pass

Validation

• ./build_workers.sh --build-and-run-tests illumination_correction — 23 passed
• ./build_workers.sh --build-and-run-tests stitched_illumination_correction — 48 passed, 1 expected BaSiCPy darkfield warning
• docker compose config --quiet
• git diff --check
• built-image inspection confirms isGPUWorker=true for both production images