get_timeinfo could be more robust

[dougiesquire]

The code for getting the frequency, start date and end date of the variables in a file was stolen and adapted very slightly from the cosima cookbook. It works okay, but it could probably be more robust.

As a example, there're data in some of the COSIMA experiments (e.g. some of the files for the wdet100 variable in 01deg_jra55v13_ryf9091) that comprise a single timestep and do not include time_bounds. These get flagged as fixed frequency fx even though they are actually 1mon. In this case, I think this is actually more of a data issue - I don't know of any way to get the frequency in this case without trying to parse it from the (non-standardised) filename (which I don't think we want to do).

[dougiesquire]

Note since #91, the filename is also parsed for frequency information. If the frequency determined from the filename and file contents differ, a warning is thrown and:

• if the contents frequency is fx, the filename frequency is used
• otherwise the contents frequency is used