BUG: signal: Division by zero in lombscargle

[WarrenWeckesser]

This problem was reported on stackoverflow: http://stackoverflow.com/questions/24702807/lomb-scargle-vs-fft-power-spectrum-crashes-with-evenly-spaced-data/

Here's a self-contained example:

In [2]: scipy.__version__
Out[2]: '0.14.0'

In [3]: from scipy.signal import lombscargle

In [4]: t = np.linspace(0, 10, 1000, endpoint=False)

In [5]: x = np.sin(4*t)

In [6]: f = np.linspace(0, 50, 500, endpoint=False) + 0.1

In [7]: q = lombscargle(t, x, f*2*np.pi)
---------------------------------------------------------------------------
ZeroDivisionError                         Traceback (most recent call last)
<ipython-input-7-119dba45b118> in <module>()
----> 1 q = lombscargle(t, x, f*2*np.pi)

/Users/warren/anaconda/lib/python2.7/site-packages/scipy/signal/_spectral.so in scipy.signal._spectral.lombscargle (scipy/signal/_spectral.c:1662)()

ZeroDivisionError: float division

The problem is that lombscargle includes the calculation atan(2 * cs / (cc - ss)), but it is possible for cc - ss to be 0.

The fix is easy: replace atan(2 * cs / (cc - ss)) with atan2(2 * cs, cc - ss).