Set up computations for quality of MD runs

[WardLT]

We need a quick way of determining whether an on-the-fly learning run recreates the correct dynamics:

• Energy averages, which could be used in heat capacity or other equation-of-state fitting
• Generate the various (P)RDFs to ensure the atomic structure is correct
• Compute the diffusivity

[miketynes]

• For NPT runs, look at the volume

[miketynes]

Once we have diffusivity, might be good to see how blending affects the transition state