Port ThermoML parsing

[lilyminium]

We need to keep the ThermoML dataset downloading/conversion functionality that we have in Evaluator. Generally this should remain the same, although preferably:

• the ID of the dataset won't be a random uuid but instead a meaningful hash
• allow properties to not contain temperature or pressure. We changed this with good intentions but I think this is now getting rid of all dhvap data
• There would be progress bars
• Ideally keeping a compatible CSV format to what we currently use
• output type: https://github.com/openforcefield/descent/blob/7cd8062e2ff222047dfaccc6e45facf614abe9db/descent/targets/thermo.py#L63 actually, we may need a new type -- see New thermoml dataset entry type? #11

This doesn't have to live here, we could have a separate data repo.

[mattwthompson]

Do we want to have something similar to openff.evaluator.properties? This is a different layer than storing ThermoML reference data but it all ends up getting mapped there. This line is the choke point, albeit very deep in the code and hard to understand at face value.

[lilyminium]

I'm not sure I understand what you mean or are referencing by having something "similar" to openff.evaluator.properties -- happy to discuss in a call if I have the wrong end of the stick here. From my perspective I'm open to having property classes (as in the skeleton PR), but IMO all the generalisation in Evaluator's ThermoML adds a lot of really confusing abstraction and I find unnecessary, since AFAICT Evaluator only ever uses _default_mapping for the natively supported properties.

[mattwthompson]

That's basically it - when I drop into a debugger when that method is called, I see two classes that aren't obviously doing different things:

ipdb> l
   2223
   2224                 mapped_property = registered_plugin.conversion_function(
   2225                     measured_property
   2226                 )
   2227                 import ipdb; ipdb.set_trace()
-> 2228                 mapped_property.source = source
   2229                 return_value.add_properties(mapped_property)
   2230         return return_value

ipdb> measured_property
<ThermoMLProperty Molar enthalpy of vaporization or sublimation, kJ/mol, substance=Fc1ccccc1Cl{solv}{x=1.000000}, thermodynamic_state={self.thermodynamic_state}, value=44.0 kilojoule / mole, uncertainty=0.6 kilojoule / mole >
ipdb> mapped_property
<openff.evaluator.properties.enthalpy.EnthalpyOfVaporization object at 0x108d4fc80>

My current work is incomplete but I'm hoping that this complexity can be simplified by moving away from a plugin architecture