fix: handle null sub-arrays in flatten

[michaelmitchell-bit]

Which issue does this PR close?

Closes #4788.

Rationale for this change

Spark flatten returns NULL for an output row when any nested sub-array in that row is NULL. Comet was using DataFusion's flatten, which preserves only the outer array null bitmap and can return incorrect values when an inner sub-array is null.

What changes are included in this PR?

This adds a Spark-compatible flatten implementation for Comet that preserves DataFusion's offset/value flattening while computing Spark-compatible output nulls. It also registers the new UDF and extends the existing flatten.sql coverage to compare Comet against Spark for null sub-array cases.

How are these changes tested?

Ran:

cargo test -p datafusion-comet-spark-expr array_funcs::flatten
cargo build --release
./mvnw test -Prelease -Dtest=none -Dsuites="org.apache.comet.CometSqlFileTestSuite flatten" -Dscalastyle.skip=true