Bug in date_range with semi-open interval and freq='B'?

[dhirschfeld]

#5637 has some examples as well

When the end date in date_range is not a valid business day it appears that date_range first sets it to a valid date before deciding whether to return the last date depending on whether the interval is closed or open on the right. This results in the date range being incorrectly truncated when the interval is open-right - e.g.

In [22]: for d in pd.date_range('20-Mar-2014', '23-Mar-2014', freq='D'):
    ...:     print d.strftime('%a %d-%b-%Y')
    ...:     
Thu 20-Mar-2014
Fri 21-Mar-2014
Sat 22-Mar-2014
Sun 23-Mar-2014

In [23]: pd.date_range('20-Mar-2014', '23-Mar-2014', freq='B')
Out[23]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-03-20, 2014-03-21]
Length: 2, Freq: B, Timezone: None

In [24]: # should return all valid business days <= '23-Mar-2014'
    ...: pd.date_range('20-Mar-2014', '23-Mar-2014', freq='B', closed='left')
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-03-20]
Length: 1, Freq: B, Timezone: None

[jreback]

related / dupe of #5637

See the comment here, I think this is somewhat arbitrary because it first needs to move to a valid business day.
https://github.com/pydata/pandas/blob/master/pandas/tseries/offsets.py#L69

I think this should be more deterministic. If you can come up with some 'rules' then I think could be straightforward.