Add in the GpuArrayFilter command

[revans2]

this fixes #10760

I still need to do some performance comparisons/testing, but functionality wise it looks good. Will update it shortly.

[revans2]

build

[revans2]

For performance numbers I ran the following query on Spark 3.4.2

spark.time(spark.range(0, 10000000000L, 1, 512).selectExpr("array(id, id + 1, id - 100, id + 3, id + 2) as ar").selectExpr("filter(ar, f -> f % 3 == 0) as f").selectExpr("SUM(size(f)) as r").show())

On an a6000 running with 4 concurrent It took 36 seconds in the worst case, I ran it 3 times and the best was about 33 seconds.

On a 32-core/64-thread Threadripper PRO 5975WX, using all 64 cores it took 688 seconds (and I didn't really want to run it again).

That is a speedup of about 19x.

[revans2]

build

[abellina]

LGTM other than copyright/indenting that @jlowe pointed out.

[revans2]

@jlowe @abellina please take another look

[revans2]

build

[revans2]

The link check is failing because of #10768