New filter that behaves with more nuance and can seed properties

[lilyminium]

I think for our ultimate training set for a new vdW fit we want:

• to incorporate some common solvents and mixtures, e.g. CO+O / CCO + O
• not limit our dhmix data by overlap with density data, as the former is so rare (related to A filter that grabs all dhmix data, then selects other (e.g. density) data with the same compounds, then fills out more density data as needed #738)
• still use diverse selection to get broad sampling across a variety of environments

This is extremely hard to achieve with current filters and may need a new one. I'm not certain this really needs to live in Evaluator all, perhaps this is better as a script, but making an issue to think about it for now.