Result mismatch with Spark in statistical aggregation function when DivideByZero occurs during expression evaluation

[NEUpanning]

Bug description

In Spark, if the configuration setting spark.sql.legacy.statisticalAggregate is set to true, statistical aggregation function will return Double.NaN instead of NULL when DivideByZero occurs during expression evaluation. For example, stddev(2.0) will return NaN in Spark but return NULL in Velox. Statistical aggregation function includes stddev, stddev_samp, variance, var_samp, skewness, kurtosis, covar_samp, corr.

Spark PR: apache/spark#29983

System information

/

Relevant logs