Some fixes for cudaCoord (group-to-group) #3269

Job	Run time
codecheck / codecheck	43m 39s
conda / conda (ubuntu-20.04)	11m 17s
docker / docker (fedora39)	19m 25s
conda / conda (macos-13)	25m 19s
linux / linux (-doc-mpi-)	40m 20s
docker / docker (rocky8)	22m 58s
mac / macports (macos-13)	23m 53s
conda / conda (macos-14)	7m 1s
linux / linux (-mpi-)	52m 24s
mac / macports (+allmodules, macos-13)	33m 27s
linux / linux (-coverage-mpi-)	1h 3m 28s
linux / linux (-debug-)	30m 18s
mac / macsimple (macos-13)	18m 4s
linux / linux (-debug-mpi-)	41m 14s
mac / macsimple (macos-14)	4m 20s
linux / linux (-intel-)	44m 54s
linux / linux (-pycv-mpi-)	19m 4s
linux / linux (-nvhpc-mpi-)	55m 40s
	9h 16m 45s

Provide feedback