Bug in 'chop.threshold' option

[jseabold]

xref #6842

Columns where every value is below the threshold disappear. From ML.

In [78]: df.as_matrix()
Out[78]:
array([[  1.00000000e+01,   7.59062963e-10],
    [  2.00000000e+01,  -1.47734680e-11],
    [  3.00000000e+01,   2.18850713e-09],
    [  4.00000000e+01,  -1.86628720e-11]])

In [79]: np.set_printoptions(suppress=True)

In [80]: df.as_matrix()
Out[80]:
array([[ 10.,   0.],
    [ 20.,  -0.],
    [ 30.,   0.],
    [ 40.,  -0.]])


From the docstring, it appears that the 'chop_threshold' option can be
used to do the same thing (and with a threshold as a parameter), but
in this case, it does something weird (perhaps a bug?):

In [81]: pd.set_option('chop_threshold', 1e-8)

In [82]: df
Out[82]:
    0  1
0  10
1  20
2  30
3  40

[dsm054]

In 0.20.2, this seems to work the way you'd expect:

In [10]: pd.set_option("chop_threshold", 1e-8)

In [11]: df
Out[11]: 
      0             1
0  10.0  0.000000e+00
1  20.0  0.000000e+00
2  30.0  0.000000e+00
3  40.0  0.000000e+00

In [12]: pd.set_option("chop_threshold", 1e-10)

In [13]: df
Out[13]: 
      0             1
0  10.0  7.590630e-10
1  20.0  0.000000e+00
2  30.0  2.188507e-09
3  40.0  0.000000e+00

[jreback]

@dsm054 can you provide a confirming test PR.