Upgrade NCCL to 2.26.2 in Kubeflow CUDA training image

Author: astefanutti

images/runtime/training/cuda/Dockerfile

@@ -83,6 +83,9 @@ RUN micropipenv install -- --no-cache-dir && \
 RUN pip install wheel
 RUN pip install --no-cache-dir flash-attn==2.7.4.post1 --no-build-isolation
 
+# Upgrade NCCL to a more recent version until we upgrade torch
+RUN pip install nvidia-nccl-cu12==2.26.2
+
 # Restore user workspace
 USER 1001