Avoid negative zero for results of decimal arithmetic operations.
Similarly for conversions from float/string to decimal.
This is a folloup patch to
Bug#18411494 WRONG COMPARSION ON BIG DECIMAL VALUES