Include basic stats about study definition in standard log output

[evansd]

In the first instance, this can be as simple as a count of the number of variables in the study definition. As long as it gets written to stdout it will end up in the logs.

The aim is for these to be machine readable, so it should use some easily greppable prefix and have a simple easily parsable syntax. Something like (though this is just an initial suggestion):

cohortextractor-stats: variable-count=123

The idea is to make it easier to debug and prevent potential performance problems by surfacing this information in a machine readable fashion.

[sebbacon]

Just to record a couple of ideas I'd had about where to log which I'd noted down last week - there may be better places:

• len(output_columns) from here
• min and max dates from here

We will also want to extract total running time stats, which we actually already log (e.g. the start time is here, not sure where the end time is); there's also other bits we could time, including temporary table downloads, total cohort generation time, time per index_date, etc. We could either log durations, or start and end times. We should probably add the same grep prefix to those log entries.