Rolling rework
Clarified and cleaned up the Rolling. Now we can have backwards rolling and step-size rolling. Also passing a mask where to skip calculation is possible. This also fixes the ramp-up and ramp-down issues for variable windows (eg. window='4d').
all pandas.rolling native features are implemented and a Rolling() bject is returned as in pandas.
# pandas
r = s.rolling(window='4d')
r.sum()
r.apply(fancyFunc)
# custom
r = customRoller(s, window='4d')
r.sum()
r.apply(fancyFunc)