With cloud nodes, favour using resources of nodes already booted first before starting other nodes

[mboisson]

When using pool nodes, Slurm will prefer to launch new nodes rather than to use available resources on existing nodes. The desired behaviour would be that it maximizes the use of nodes that are online before launching new ones.

[cmd-ntrf]

We have identified that to be able to achieve this, we have to patch Slurm. This lead to the creation of this subproject:
https://github.com/MagicCastle/slurm-select-cons_tres_cloud

Patches will be applied to Slurm source code when compiling RPMs on COPR:
https://copr.fedorainfracloud.org/coprs/cmdntrf/

Integration of the patch the feature will be limited to Slurm >= 25.

The missing piece is allowing the user to enable the feature via puppet (i.e.: adding PreferPoweredUpNodes to SelectTypeParameters).