add 'nrel' spa option for ET irradiance, fix doy - 1 bug

[wholmgren]

This PR adds a new 'nrel' option to irradiance.extraradiation that uses the well tested ephemeris code in spa.py. It also fixes the doy - 1 bug described in #211.

My plan is to let #214 sit for another day for comments, merge it, and then rebase/merge this PR.

[wholmgren]

Here are some benchmarks:

times = pd.DatetimeIndex(start='2015', end='2016', freq='1min')

In [40]: %timeit pvlib.irradiance.extraradiation(times, method='nrel', how='numba')
1 loop, best of 3: 1.67 s per loop

In [41]: %timeit pvlib.irradiance.extraradiation(times, method='nrel')
1 loop, best of 3: 3.87 s per loop

In [42]: %timeit pvlib.irradiance.extraradiation(times, method='pyephem')
1 loop, best of 3: 8.34 s per loop

In [43]: %timeit pvlib.irradiance.extraradiation(times, method='spencer')
10 loops, best of 3: 45.8 ms per loop

In [44]: %timeit pvlib.irradiance.extraradiation(times, method='asce')
10 loops, best of 3: 26 ms per loop

I was hoping that the nrel method would be faster, but it's still an improvement over pyephem.

[wholmgren]

A sloppy copy/paste job led to an unnecessary slow line of code in solarposition.nrel_earthsun_distance. I removed the slow line and reran the tests:

In [61]: times = pd.DatetimeIndex(start='2015', end='2016', freq='1min')

In [62]: %timeit pvlib.irradiance.extraradiation(times, method='nrel', how='numba')
The slowest run took 10.33 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3: 215 ms per loop

In [63]: %timeit pvlib.irradiance.extraradiation(times, method='nrel')
1 loop, best of 3: 953 ms per loop

In [64]: %timeit pvlib.irradiance.extraradiation(times, method='pyephem')
1 loop, best of 3: 7.9 s per loop

In [65]: %timeit pvlib.irradiance.extraradiation(times, method='spencer')
10 loops, best of 3: 64.4 ms per loop

In [66]: %timeit pvlib.irradiance.extraradiation(times, method='asce')
10 loops, best of 3: 28.5 ms per loop