more Timer functionality

[sbailey]

This PR adds several features that I wanted when actually putting it into use in desispec:

• silent option: record timing, but don't print log messages, e.g. to record timing for multiple MPI ranks without having every rank blasting timing log messages.
• calculate_stats (like Ted's example from CMB-land suggested in the original Timer PR add Timer class to standardize algorithm timing reports #151): combine multiple Timers into a report with min/max/mean/median of start/stop/duration.
• options to start and stop to set the time to use instead of time.time(), useful for recording timing of events that may have occurred before the timer was created, e.g. python startup and imports.
• Timer.cancel to discard a named timer that had been started.

I'll submit a companion desispec PR that uses these features, but this PR will need to be reviewed and merged before that one can be finalized.

[sbailey]

Travis tests are catching a timezone parsing issue that doesn't arise at NERSC or on my laptop. I'll investigate what the difference is, and back off on fancy date string parsing if needed. There might be some Travis churn as I sort this out.

[weaverba137]

You're using pytz right?

[sbailey]

No I wasn't using pytz (only vaguely aware of it...) Maybe for now I will just drop support for the fully generic Unix date output and require date +%s or ISO-8601 (which itself has lots of options...). I can return to adding vanilla date support if the pain of not having it exceeds the pain of implementing it.

[weaverba137]

OK, ready for me to take a look?

[sbailey]

Yes, please. Thanks.

[weaverba137]

Overall looks fine, but please note the comments. Most importantly, make sure self._print() is used consistently throughout the Timer class.

[weaverba137]

PS, please update the change log.

[weaverba137]

OK to merge assuming tests pass.