[BUG] Sum and multiply aggregations promote unsigned input types to a signed output

**Describe the bug**
When performing aggregations, the output types are often upscaled to help combat overflow situations. For example, performing a sum aggregation on an INT32 column will produce an INT64 result. However performing a sum aggregation on a UINT32 column produces an INT64 result rather than a UINT64 result.

**Steps/Code to reproduce bug**
Perform a sum aggregation with an input column of UINT32 and note that the result is INT64. Here's a snippet of a session doing this with the cudf Java API in the Spark REPL shell:
```
scala> import ai.rapids.cudf._
import ai.rapids.cudf._

scala> val t = new Table(ColumnVector.fromInts(0), ColumnVector.fromUnsignedInts(0))
t: ai.rapids.cudf.Table = Table{columns=[ColumnVector{rows=1, type=INT32, nullCount=Optional[0], offHeap=(ID: 5 7feec1b5ac90)}, ColumnVector{rows=1, type=UINT32, nullCount=Optional[0], offHeap=(ID: 9 7feec1b5a280)}], cudfTable=140663428850592, rows=1}

scala> t.groupBy(0).aggregate(GroupByAggregation.sum().onColumn(1))
res0: ai.rapids.cudf.Table = Table{columns=[ColumnVector{rows=1, type=INT32, nullCount=Optional.empty, offHeap=(ID: 10 7ff0b7039860)}, ColumnVector{rows=1, type=INT64, nullCount=Optional.empty, offHeap=(ID: 11 7ff0b7039760)}], cudfTable=140671839344304, rows=1}
```

**Expected behavior**
Unsigned input types should be promoted to unsigned output types for any aggregations where the sign of the result cannot change for unsigned inputs (e.g.: sum and multiply)

**Additional context**
See @jrhemstad's comment at https://github.com/rapidsai/cudf/issues/10102#issuecomment-1023502832