Optimizing the performance by reusing the dense `select_k` #4432

Job	Run time
checks / check-style	1m 55s
checks / other-checks	37s
devcontainer / build (amd64, 12.2, conda)	3m 55s
devcontainer / build (amd64, 12.2, pip)	3m 54s
wheel-build-pylibraft / compute-matrix	7s
conda-cpp-build / compute-matrix	3s
conda-cpp-build / 11.8.0, 3.10, amd64, ubuntu22.04	1m 39s
conda-cpp-build / 12.2.2, 3.10, amd64, ubuntu22.04	1m 40s
conda-cpp-build / 11.8.0, 3.10, arm64, ubuntu22.04	1m 42s
conda-cpp-build / 12.2.2, 3.10, arm64, ubuntu22.04	1m 47s
wheel-build-pylibraft / 11.8.0, 3.9, amd64, centos7	1m 33s
wheel-build-pylibraft / 12.2.2, 3.9, amd64, centos7	1m 37s
wheel-build-pylibraft / 11.8.0, 3.10, amd64, centos7	1m 40s
wheel-build-pylibraft / 12.2.2, 3.10, amd64, centos7	1m 38s
wheel-build-pylibraft / 11.8.0, 3.11, amd64, centos7	1m 39s
wheel-build-pylibraft / 12.2.2, 3.11, amd64, centos7	1m 39s
wheel-build-pylibraft / 11.8.0, 3.9, arm64, rockylinux8	1m 41s
wheel-build-pylibraft / 12.2.2, 3.9, arm64, rockylinux8	1m 35s
wheel-build-pylibraft / 11.8.0, 3.10, arm64, rockylinux8	1m 40s
wheel-build-pylibraft / 12.2.2, 3.10, arm64, rockylinux8	1m 39s
wheel-build-pylibraft / 11.8.0, 3.11, arm64, rockylinux8	1m 40s
wheel-build-pylibraft / 12.2.2, 3.11, arm64, rockylinux8	1m 39s
wheel-tests-pylibraft	1s
wheel-build-raft-dask	1s
wheel-tests-raft-dask	1s
conda-cpp-tests	1s
conda-python-build	1s
conda-cpp-checks	1s
conda-python-tests	1s
docs-build	1s
pr-builder	1s
	37m 8s

Provide feedback