Fix parsing of binary floating point number