Fixes #687.

The docstring for the _default_frequency_range() function says

This code looks at the poles and zeros of all of the systems that
we are plotting and sets the frequency range to be one decade above
and below the min and max feature frequencies, rounded to the nearest
integer.

The old code rounded to the floor and ceiling of the range, which occasionally adds another decade to one or the other end of the frequency range if the feature falls just inside of the outermost decade.

The new code rounds to the nearest integer, so it only adds half a decade at minimum.

Coverage increased (+0.09%) to 93.591% when pulling 021372f on bnavigator:omega_default_round into a111b03 on python-control:master.

Looking at the coveralls report, it turns out the Hz parameter was never passed through to _default_frequency_range(). Fixed now.

From the sisotool test which had to be changed:

from numpy import array
from matplotlib import pyplot as plt
from control import TransferFunction, bode_plot

tsys = TransferFunction(array([1000]), array([  1,  25, 100,   0]))
# poles: [-20.,  -5.,   0.]

for Hz in [True, False]:
    plt.figure()
    bode_plot(tsys, Hz=Hz)

Hz	False	True
Before
After

@bnavigator In thinking through this overnight, I wonder if we might want to revert to the old behavior (floor, ceil) and update the documentation to match (+ update the unit tests to avoid the corner case). I'm worried that only going out a half decade might now show the asymptotic frequency response. I'll work up a couple of examples and post an issue so that we can discuss.

Hmm, okay. We need to find another way to make test_nyquist_indent more robust then.