PERF: infer_datetime_format fails without zero-padding

[chris-b1]

I think this can be fixed by just padding tokens in the inference step. The dates do parse, just always fall back to the slow path, even with infer_datetime_format=True

In [22]: from pandas.tseries.tools import _guess_datetime_format

In [23]: _guess_datetime_format('02/02/2014')
Out[23]: '%m/%d/%Y'

In [24]: _guess_datetime_format('2/2/2014')

[jreback]

I think you can add additional formats here.