Fixed bug in NLTE radiative transfer

[ThomasCeulemans]

In lower density regimes, population inversionsmight occur. If this happens, the opacity will become negative, leading to negative source functions and optical depths. The Feautrier solver cannot handle this. This is why we bound the opacities from below.

The actual change is just bounding the line opacities at the correct place in the code (+adding an extra parameter for bounding the line opacity and adding warnings when the line opacity becomes negative).

[ThomasCeulemans]

This current build passes the test, but I am not confident that the NLTE radiative transfer does not show numerical errors (as the resulting images change depending on the hyperparameter values).

[ThomasCeulemans]

This solution does not seem to work, closing. I will replace it by manually resetting masering level populations to LTE.