BUG: Unable to infer negative freq

[sinhrks]

Fixed 2 issues:

• date_range can't handle negative calendar-based freq (like A, Q and M).

import pandas as pd

# OK
pd.date_range('2011-01-01', freq='-1D', periods=3)
# DatetimeIndex(['2011-01-01', '2010-12-31', '2010-12-30'], dtype='datetime64[ns]', freq='-1D')

# NG
pd.date_range('2011-01-01', freq='-1M', periods=3)
# DatetimeIndex([], dtype='datetime64[ns]', freq='-1M')

• Unable to infer negative freq.

pd.DatetimeIndex(['2011-01-05', '2011-01-03', '2011-01-01'], freq='infer')
# DatetimeIndex(['2011-01-05', '2011-01-03', '2011-01-01'], dtype='datetime64[ns]', freq=None)

pd.TimedeltaIndex(['-1 days', '-3 days', '-5 days'], freq='infer')
# TimedeltaIndex(['-1 days', '-3 days', '-5 days'], dtype='timedelta64[ns]', freq=None)

[jreback]

thanks @sinhrks