FIX: Correct output of mlab._spectral_helper when scale_by_freq=False

[e-q]

Closes #4328

[tacaswell]

This causes all of the spectral tests to fail.

[e-q]

True, didn't check the tests before submitting. In this case, the tests are in error.

Specifically, all of the failures are in test_psd_windowarray_scale_by_freq, which tests assert_allclose(spec_s, spec_n/self.Fs, atol=1e-08), where spec_s is computed with scale_by_freq=True and spec_n is computed with scale_by_freq=False. This equality is only true when using a flat window.

I've chosen to modify the tests to use a flat window, instead of the function's default, which I believe is a Hanning window. If, instead, we want to test with the Hanning window, the test can be modified to calculate the difference in window correction factors.

The tests now seem to pass on my machine, let's see what Travis thinks...

[e-q]

Hmm, now the tests are failing by a factor of 100. I'll get back to this soon.

[e-q]

There we go. I used the hanning window after all.