Improve aggregation performance of average on DECIMAL128 columns [databricks]

[jlowe]

Fixes #4722.

Accelerates DECIMAL128 average aggregations using the same technique employed in #4688. This allows the use of the cudf hash-based algorithm instead of the slower sort-based algorithm for these types of aggregations.

[jlowe]

build

[jlowe]

CI failed due to recent cudf string split API breakage:

Error: :43:11.360Z] [ERROR] [Error] /home/jenkins/agent/workspace/jenkins-rapids_premerge-github-3972/sql-plugin/src/main/scala/org/apache/spark/sql/rapids/stringFunctions.scala:1371: overloaded method value stringSplitRecord with alternatives:
[2022-02-14T17:43:11.361Z]   (x$1: String,x$2: Int)ai.rapids.cudf.ColumnVector <and>
[2022-02-14T17:43:11.361Z]   (x$1: String,x$2: Boolean)ai.rapids.cudf.ColumnVector
[2022-02-14T17:43:11.361Z]  cannot be applied to (ai.rapids.cudf.Scalar, Int)

[jlowe]

build

[jlowe]

CI couldn't find resources:

[2022-02-15T15:28:14.617Z] Waited 60 times already, stopping

[jlowe]

build