[BUG] Inner Join dropping data with bucketed Table input

[tgravescs]

Describe the bug

Inner join is losing data when one of the tables is written bucketed. See example below for the specifics.

With code below results is empty:
+---+--------+---+-----------+
|Url|SatCount|Url|ResolvedUrl|
+---+--------+---+-----------+
+---+--------+---+-----------+

It should be:
+--------------------------------------+--------+--------------------------------------+---------------------------------------+
|Url |SatCount|Url |ResolvedUrl |
+--------------------------------------+--------+--------------------------------------+---------------------------------------+
|http://fooblog.com/blog-entry-116.html|21 |http://fooblog.com/blog-entry-116.html|https://fooblog.com/blog-entry-116.html|
|http://fooblog.com/blog-entry-116.html|21 |http://fooblog.com/blog-entry-116.html|http://fooblog.com/blog-entry-116.html |
+--------------------------------------+--------+--------------------------------------+---------------------------------------+

Note that if I turn off the GPU shuffle (spark.conf.set("spark.rapids.sql.exec.ShuffleExchangeExec", "false")) then it works and produces the correct results. I had tried changing the join to be a CPU join but it still failed which makes me think its not the join itself but something higher up. Also if you throw in an extra repartition after reading from the table then it works ok as well.

Steps/Code to reproduce bug

$SPARK_HOME/bin/spark-shell --master local --jars ~/.m2/repository/ai/rapids/cudf/0.15-SNAPSHOT/cudf-0.15-SNAPSHOT-cuda10-1.jar,/home/tgraves/workspace/spark-rapids-another/dist/target/rapids-4-spark_2.12-0.2.0-SNAPSHOT.jar --conf spark.driver.extraJavaOptions=-Duser.timezone=GMT --conf spark.sql.session.timeZone=UTC --conf spark.executor.extraJavaOptions=-Duser.timezone=GMT --conf spark.plugins=com.nvidia.spark.SQLPlugin --conf spark.rapids.sql.explain="NOT_ON_GPU"

spark.conf.set("spark.sql.autoBroadcastJoinThreshold", "-1")

val rdd = sc.parallelize(Seq(("http://fooblog.com/blog-entry-116.html", "https://fooblog.com/blog-entry-116.html"), ("http://fooblog.com/blog-entry-116.html", "http://fooblog.com/blog-entry-116.html")))

val resolved = rdd.toDF("Url", "ResolvedUrl")
val rdd2 =  sc.parallelize(Seq(("http://fooblog.com/blog-entry-116.html", "21")))

val feature = rdd2.toDF("Url", "SatCount")
feature.write
      .bucketBy(4000, "Url")
      .sortBy("Url")
      .format("parquet")
   .mode("overwrite")
      .option("path", "/home/tgraves/tgravesfeatureset")
      .saveAsTable("tgravesfeatureset")

val testurls = spark.sql("SELECT Url, SatCount FROM tgravesfeatureset")
val res = testurls.join(resolved, testurls("Url") === resolved("Url"), "inner")
res.show(false)

[tgravescs]

So it seems we should have been not allowing this to run on the GPU due to the hashing won't be the same across cpu and gpu.