(very) low-temperatures cause NaNs in likelihood

[benburningham]

It appears that having a profile that gets much cooler than the 75K minimum of the opacity temperature grid (e.g. 20K) can lead to NaNs in the likelihood. Possibly due to bad extrapolation of opacities beyond the grid.

Probably needs further testing to pin down how & when this happens.

Solutions:

1. set prior to restrict T to above 75K?
2. don't extrapolate, just appy 75K opacities to lower temperature?

Not sure....