Accommodate altered semantics of cudf::lists::contains()

[mythrocks]

Depends on rapidsai/cudf/pull/9510.
Depends on rapidsai/cudf/pull/9901.

spark-rapids relies on libcudf's cudf::lists::contains() for its implementation of array_contains().

rapidsai/cudf/pull/9510 changes the semantics of lists::contains(), with regard to rows containing nulls. Specifically, if a list row contains at least one null element, and is found not to contain the search key, libcudf will now return false instead of null.
SparkSQL expects to return null in those cases.

This commit accommodates the change in libcudf's semantics, to keep its own existing behaviour.

[mythrocks]

build

[mythrocks]

This PR is ready for review, although it's still in draft. It can't be merged until the dependencies in CUDF have been merged.

[mythrocks]

(Modified for changed CUDF function name. copyWithValidity -> copyWithBooleanColumnAsValidity.)

[mythrocks]

Re-tested after changes to dependencies in libcudf. Will rebuild after rapidsai/cudf/pull/9510 is merged.

[revans2]

Just a few minor nits

[mythrocks]

rapidsai/cudf/pull/9510 was just committed. I've pulled this PR out of draft.

[revans2]

build

[revans2]

Even though rapidsai/cudf/pull/9510 has not gone all the way through CI yet. I think we can still test/commit this. Because the extra processing should be a noop if listContains is behaving the old way too.

[revans2]

oops I was wrong. It needs a new API that was missing before.

[tgravescs]

build

[mythrocks]

Merging this to head off impending CI breakage.

[mythrocks]

Thank you for the review, @revans2. @tgravescs, thank you for confirming that the CI has passed with the new libcudf artifact.