Population bug fixes for Python 80 release (used for Matthews+ 16 Quasar paper)

[jhmatthews]

This introduces some more robust error checking and some more stable code to deal with situations where Python is struggling to solve correctly for matom level populations when the estimators are noisy.

In particular, large changes include:

• actually using the bound-free estimators above 1.2e16 Hz (up to 5e18 Hz), instead of adopting a dilute approximation. This was a relic of BBody SEDs only which caused us to get the Helium ionization state wrong.
• a function get_dilute_estimators() which populates estimators with dilute approximations IF the level populations fail in macro_pops().
• a function check_stimulated_recomb() which checks each cycle if the stimulated recomb will be higher than ionization rate in a given cell. This can happen in unconverged hot cells where T_R >> T_E. In the actual matom we set to 0 and hope spontaneous recombiantion takes care of things, and report errors to the user (but just one error per cell).

Smaller changes:

• check the determinant of the rate matrix is not 0
• add a check for stimfac < 0 as well as > 1