EoS fitting routines don't fail gracefully

[bobmyhill]

It's really common for there to be insufficient data to constrain an EoS. In such circumstances, lsq procedures drive the parameters outside the range of validity of the EoS and some bit of the EoS fails (often due to very low K').

BurnMan should deal with this more gracefully than it currently does. To do:

• add try/except to the PTp and XPTp functions in eos_fitting to catch exceptions and print a better message
• restore parameters to their initial values after failure, so that the user can use the same mineral / solution object in their next fitting attempt.