BUG: quantile ignores axis kwarg

[TomAugspurger]

Closes #7306

We may want to wait on this till later today while I figure out what will happen in #7308

[TomAugspurger]

This should be ok once Travis passes.

One odd thing is that numeric_only is always considered by column, even when axis=1.
So if you have something like

[[1,    2,  3],
 ['a', 'b', 4]]

It will exclude the first two columns since they aren't numeric. But this is the same as the previous behavior.

[TomAugspurger]

Just to expand on that last comment. As before, we get:

In [34]: df
Out[34]: 
   0  1  2
0  1  2  3
1  a  b  4

In [35]: df.quantile(.5, axis=1)
Out[35]: 
0    3
1    4
dtype: float64

So the first two columns are excluded since they're non-numeric, so we're doing quantile on just the last column:

In [38]: df[2]
Out[38]: 
0    3
1    4
Name: 2, dtype: int64

Since axis=1, we take the median of each length 1 array and get the same thing back.

What I think should happen is we transpose first, and see that the first col is numeric and calculate the quantile on that column:

In [52]: df.T
Out[52]: 
   0  1
0  1  a
1  2  b
2  3  4

So df.quantile(.5, axis=1) should be equivalent to df.T.quantlie(.5, axis=0)

But this would potentially break existing code so I'll hold off on that till the future.

[TomAugspurger]

I added one more test covering this; I'll merge when that's green.

[jreback]

gr8 looks good

[jreback]

thanks @TomAugspurger