Experiments not working with default scaling

[michaelmckinsey1]

Ref: #701 (comment)

Several experiments (hpl, kripke, remhos, etc.) check for multiple scaling options (single_node, strong, weak, throughput) and error if there are more than 1 specified. Since +single_node is silently added to every config, certain experiments require ~single_node to be explicitly specified in order to perform other scaling experiments, e.g. hpl+strong~single_node works, hpl+strong doesn't.

Disabling single_node to be on by default also breaks certain experiments, that expect +single_node to set certain experiment variables in the experiment.py.

1. Should the default behavior instead be to error if the scaling type is not specified?
2. Do we then expect the following spec syntax for every experiment: [benchmark]+[programming_model]+[scaling], e.g. amg2023+openmp+strong, otherwise experiment.py errors?

Proposed changes:

• Remove single_node
• Make scaling modes mutually exclusive

[michaelmckinsey1]

Addressed by #735