[FEA] Better precision range for decimal multiply, and possibly others

[revans2]

Is your feature request related to a problem? Please describe.
Spark does some odd things when it comes to decimal multiply as you can see from #1564 It would be really great if we could work with Apache Spark, or possibly cudf to solve this problem in some way.

Describe the solution you'd like
I see two possibilities.

1. We work with Apache Spark to make it so there is a clear and simple way tell what the original values were, or we strip out the unneeded manipulations in the first place.
2. We work with cudf so that we can combine decimal multiply, INT divide, etc so we can do the operation at a very large precision and then scale it down at the end to match what spark does.

Describe alternatives you've considered
Magic???

Additional context
None

[revans2]

This was done as a part of the decimal 128 work