Use gpuci_mamba_retry to install local artifacts. (#9951)

I see CI timeouts occurring at the step where local conda artifacts for libcudf and libcudf_kafka are installed. This PR uses `gpuci_mamba_retry` instead of `conda` to install those local artifacts (this change was also recently made in rapidsai/cugraph#1928).

Example timeouts:
- https://gpuci.gpuopenanalytics.com/job/rapidsai/job/gpuci/job/cudf/job/prb/job/cudf-gpu-test/CUDA=11.5,GPU_LABEL=driver-495,LINUX_VER=ubuntu20.04,PYTHON=3.8/5764/console
- https://gpuci.gpuopenanalytics.com/job/rapidsai/job/gpuci/job/cudf/job/prb/job/cudf-gpu-test/CUDA=11.5,GPU_LABEL=driver-495,LINUX_VER=ubuntu20.04,PYTHON=3.8/5773/console

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Ray Douglass (https://github.com/raydouglass)

URL: #9951

ci/gpu/build.sh

@@ -219,7 +219,7 @@ else
     KAFKA_CONDA_FILE=${KAFKA_CONDA_FILE//-/=} #convert to conda install
 
     gpuci_logger "Installing $CUDF_CONDA_FILE & $KAFKA_CONDA_FILE"
-    conda install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
+    gpuci_mamba_retry install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
 
     install_dask