Trim leading zeros from a floating point exponent, per C99.  See issue 1600.  As far as I know, this only affects Windows.  Add float type 'n' to PyOS_ascii_formatd (see PEP 3101 for 'n' description).