Free curand states before the thread is destroyed #3163

Job	Run time
build-and-test-cpp-x86_64 (ubuntu-22.04, dnnl)	5m 13s
build-and-test-cpp-aarch64 (openblas)	4m 39s
build-and-test-cpp-x86_64 (ubuntu-22.04, mkl)	5m 34s
check-python-style	15s
build-and-push-docker-images	27m 55s
build-python-wheels (windows-2019, auto64)	1s
build-python-wheels (macos-13, arm64)	1m 46s
build-python-wheels (ubuntu-22.04, auto64)	1m 23s
build-python-wheels (macos-13, x86_64)	1m 38s
build-python-wheels (ubuntu-22.04, aarch64)	1m 38s
test-python-wheels	0s
build-and-deploy-docs	0s
publish-python-wheels-on-pypi	-1s
	50m 1s

Provide feedback