Fix systemd unit file created by hostsetup to allow MPI jobs to use RDMA

In response to RFE #89685, requesting the creation and use of a systemd unit file, the fix 432732.

The fix modified hostsetup to create the unit file, below, and register the lsfd service.

[Unit]
Description=IBM Spectrum LSF
After=network.target nfs.service autofs.service gpfs.service

[Service]
Type=forking
ExecStart=${LSF_SERVERDIR}/lsf_daemons start
ExecStop=${LSF_SERVERDIR}/lsf_daemons stop
KillMode=none

[Install]
WantedBy=multi-user.target

Unfortunately, as systemd runs without reference to system settings - specifically /etc/security/limits.conf - when the lsfd service is started by systemd, the daemons inherit those limits from systemd and any jobs run on such hosts inherit *their* limits from sbatchd.

The upshot is that MPI jobs are unable to use RDMA, since they are unable to lock enough memory to allocate RDMA buffers, causing the jobs to fail.

eg

fluent_mpi.17.0.0: Rank 0:1: MPI_Init: ibv_create_cq() failed 4
fluent_mpi.17.0.0: Rank 0:1: MPI_Init: Can't initialize RDMA device
fluent_mpi.17.0.0: Rank 0:1: MPI_Init: Internal Error: Cannot initialize RDMA protocol

You may test if this is the cause of the problem by brunning 'ulimit -l' on the node in question.

Systemd provides a mechanism to rectify this weakness in its model. By adding the line 'LimitMEMLOCK=infinity' (as shown below) to the unit file, re-brunning the 'ulimit -l' command shows 'unlimited' and MPI jobs may proceed correctly.

[Unit]
Description=IBM Spectrum LSF
After=network.target nfs.service autofs.service gpfs.service

[Service]
LimitMEMLOCK=infinity
Type=forking
ExecStart=${LSF_SERVERDIR}/lsf_daemons start
ExecStop=${LSF_SERVERDIR}/lsf_daemons stop
KillMode=none

[Install]
WantedBy=multi-user.target