Add workflow to train SR-GNN model from ensemble of nekRS simulations

[rickybalin]

This PR adds a user-driven workflow for training the SR-GNN model within nekRS-ML on high and low fidelity simulation data. The standard approach to train the SR-GNN model runs a single nekRS simulation which projects the high-p solution (the target of the model) to low-p space (the input to the model). In some instances, the projected low-p solution is significantly different than what an actual low-p nekRS simulation would see. This workflow aims to address this discrepancy by running a series of low- and high-p nekRS simulations to generate the training data.

The workflow uses EnsembleLauncher to run an ensemble of nekRS simulation, each starting from a different .f field file produced by an initial pilot simulation. The remainder of the workflow is the same as the turbChannel_srgnn example.

To Do:

• Sketch workflow, from pilot simulation, to ensemble, to pre-processing the data, training and inference.
• Validate ensemble piece of the workflow runs as expected
• Validate pre-processing and training to make sure all data is used as expected
• Update READMEs

[rickybalin]

Two issues to fix:

• project allocation is hard coded in setup_case
• the case name is not obtained correctly because there are 2 par files in the directory