Fixes #9194

For some reason, only in cases of addition and subtraction precision gets assigned to result. Probably, this is why only in these two methods there was a cast. Now, there is a test for type of result in both cases. Also for null, as subtraction didn't check for that. Addition and subtraction with tricky types works now:

irb(main):002> BigDecimal(10) + Complex(1,2)
=> (0.11e2+2i)
irb(main):003> BigDecimal(10) - Complex(1,2)
=> (0.9e1-2i)

mult() has this:

        if (val == null) { // TODO: what about n arg?
            return callCoerced(context, sites(context).op_times, b, true);
        }

So it seems that precision may be lost or ignored in most operations. This PR does not address any of that.