What do we do about unphysical data?

This mainly applies to sesame data, but it's possible for other EOS to produce garbage states as well (see the Gruneisen EOS before I made significant modifications to protect against bad inputs).

We can do a number of different things, but they all tend to either be modifications to the result at the EOS level or some sort of production hardening layer that is applied to the EOS. My argument in favor of shifting these to a production hardening layer is that it allows the EOS to return whatever it is designed to return and nothing more. If a user wants the raw data, they can get the raw data. But if a user wants to avoid bad data (negative bulk modulus, negative heat capacity, negative temperatures, etc.) they can apply a modifier that, as the name implies, modifies the EOS output so that it is constrained to logical bounds.

Start by reviewing how the Sesame EOS and Gruneisen EOS handle invalid states, including the existing protections mentioned in the issue. Determine whether a production hardening layer is the chosen approach and define its constraints, affected EOS outputs, and validation criteria; the issue does not name files or tests.