Switch pytest-xdist algorithm to worksteal (#708)

This PR uses the `worksteal` algorithm to accelerate parallel pytests. See rapidsai/cudf#15207.

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

Approvers:
  - GALI PREM SAGAR (https://github.com/galipremsagar)
  - Jake Awe (https://github.com/AyodeAwe)

URL: #708

ci/test_python.sh

@@ -51,7 +51,7 @@ timeout 20m pytest \
   --cache-clear \
   --junitxml="${RAPIDS_TESTS_DIR}/junit-cucim.xml" \
   --numprocesses=8 \
-  --dist=loadscope \
+  --dist=worksteal \
   --cov-config=.coveragerc \
   --cov=cucim \
   --cov-report=xml:"${RAPIDS_COVERAGE_DIR}/cucim-coverage.xml" \

ci/test_wheel.sh

@@ -22,8 +22,16 @@ else
     if [[ ${CUDA_MAJOR_VERSION} == "11" ]]; then
         # Omit I/O-related tests in ./python/cucim/tests due to known CUDA bug
         # with dynamic loading of libcufile.
-        python -m pytest ./python/cucim/src/
+        python -m pytest \
+          --junitxml="${RAPIDS_TESTS_DIR}/junit-cucim.xml" \
+          --numprocesses=8 \
+          --dist=worksteal \
+          ./python/cucim/src/
     else
-        python -m pytest ./python/cucim
+        python -m pytest \
+          --junitxml="${RAPIDS_TESTS_DIR}/junit-cucim.xml" \
+          --numprocesses=8 \
+          --dist=worksteal \
+          ./python/cucim
     fi
 fi