Fix overflow behavior for spark decimal sum aggregate function

[zhli1142015]

The result / intermediate type's precision is min(38, input's precision + 10).
If it's less than 38, we should check overflow with it not 38 / 18. Now the overflow
check doesn't work as expected.
Below case repro the failure:

val df = spark.sql("select sum(if(cast(id % 10000 as decimal(9, 2)) < cast(9999999.00 as decimal(9, 2)), cast(9999999.00 as decimal(9, 2)), cast(id % 10000 as decimal(9, 2)))), count(*) from range(100000000000)")
df.collect

Null should be returned, but failure happens.

org.apache.spark.SparkRuntimeException: Error while decoding: org.apache.spark.SparkArithmeticException: [DECIMAL_PRECISION_EXCEEDS_MAX_PRECISION] Decimal precision 20 exceeds max precision 19.
createexternalrow(input[0, decimal(19,2), true].toJavaBigDecimal, staticinvoke(class java.lang.Long, ObjectType(class java.lang.Long), valueOf, input[1, bigint, false], true, false, true), StructField(sum((IF((CAST((id % 10000) AS DECIMAL(9,2)) < CAST(9999999.00 AS DECIMAL(9,2))), CAST(9999999.00 AS DECIMAL(9,2)), CAST((id % 10000) AS DECIMAL(9,2))))),DecimalType(19,2),true), StructField(count(1),LongType,false)).

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	291e366
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/6704975a94e0f10008ae04e9

[zhli1142015]

cc @rui-mo , thanks