Inconsistency in Money/Double arithmetic

[sheesania]

See #13 (comment)

[gbeeley]

Here's some research I did on this a while back:

LH       RH       subtraction      addition        notes
integer  integer  integer          integer
integer  double   double, pro LH   double, pro LH
integer  money    money, pro LH    money, pro LH
integer  *        integer, cnv RH  integer, cnv RH
double   integer  double, pro RH   double, pro RH
double   double   double           double
double   money    double           double          resulting amounts in whole currency
double   *        double, cnv RH   double, cnv RH
money    integer  money            money           RH treated as whole currency
money    double   double, cnv LH   money, cnv RH   *inconsistency*
money    money    money            money
money    *        money, cnv RH    ?
string   *        string, cnv RH   string, cnv RH  convert to string before performing operation```

[gbeeley]

I should add here -- the research I had done indicated that among other SQLs it is standard for the type precedence to be (highest to lowest): Fixed point (money), floating point (double), and then integer. This would change several of the above result types. With multiplication and division, the current code follows dimensional analysis logic where possible (so, for example, money × money is not defined, and money ÷ money is double or integer).