solarposition.hour_angle is inefficient and fails on some pandas versions

[wholmgren]

Describe the bug
For some reason related to timezone handling, the solarposition.hour_angle function fails on pandas versions 0.15-0.18. We test 0.14 and latest. Apparently we were on pandas 0.19 or greater when this function was created.

Also, the loop over datetime values is probably slow.

    hours = np.array([(t - t.tz.localize(
        dt.datetime(t.year, t.month, t.day)
    )).total_seconds() / 3600. for t in times])

I think this gives the same result:

hours = (times.astype(int) - times.normalize().astype(int)) / 1e9 / 3600

Versions:

• pvlib.__version__: 0.6.0
• pandas.__version__:
• python:

Additional context
probably should address #597 first

[mikofski]

@wholmgren I believe this may be related to the pandas issues I observed in #583 - I noticed this hour_angle code, and I thought I should have addressed it then, but I didn't. It was my bad code from the Bird clear sky model, sorry!

[wholmgren]

No worries, Mark. Do you want to address this and #597 in #583? Or do you think we should handle them separately?

[mikofski]

I will address them, but I'd rather open a new PR to fix the hour_angle issues, unless you would prefer not

[wholmgren]

New PR sounds good to me. Thanks!