Text spectra have too low precision

[alexji]

By default I only save 4 digits of precision when writing out spectrum text files. This does not work when ivar is too small, and fails often if saving out spectra with counts instead of normalized (not used in SMHR, but yes elsewhere).

The best thing to do is to change to save as .fits instead of .txt in session.py, but I have a note that this is sometimes broken.