[FEA] check decimal inputs before multiply and divide operations

[revans2]

Is your feature request related to a problem? Please describe.
This is another one of my crazy ideas. I have not verified anything with it to see how performant it might be. We know that multiply and divide are very expensive to do as decimal 128 values. But we also know that Spark often will make values grow a lot larger than they need to be. It would be great if we could come up with a way to very cheaply check if what is stored in a column needs 128-bits or if we could do the computation as 64-bits instead. We might be able to use a min/max reduction on the column vector to detect this.

[ttnghia]

How about binop comparison? Comparing all elements to the int64_t::max and do a reduction to see if all the numbers are fit within 64 bit rep.