[BUG] GpuJsonScan incorrect behavior when parsing dates

[andygrove]

Describe the bug
Reading a JSON file containing dates in yyyy-MM-dd format and when specifying the option dateFormat as dd/MM/yyyy causes a runtime exception of One or more values is not a valid date when running on the GPU.

Steps/Code to reproduce bug

import org.apache.spark.sql.types._
import org.apache.spark.sql.types.DataTypes._

val schema = StructType(Seq(StructField("number", DateType, false)))

val df = spark.read.option("dateFormat","dd/MM/yyyy").schema(schema).json("integration_tests/src/test/resources/dates.json")

df.show(false)

CPU Result

+----------+
|number    |
+----------+
|2020-09-16|
|2020-09-16|
|2020-09-16|
|1581-01-01|
|1583-01-01|
+----------+

GPU Result

java.time.DateTimeException: One or more values is not a valid date
	at com.nvidia.spark.rapids.GpuTextBasedDateUtils$.$anonfun$castStringToDate$3(GpuTextBasedPartitionReader.scala:762)

Expected behavior

Environment details (please complete the following information)

Additional context

[andygrove]

Here is an integration test that highlights some of the issues:

@pytest.mark.parametrize('data_gen', [
    StringGen('[1-9]{4}-[1-3]{1,2}-[1-3]{1,2}', nullable=False),
    StringGen('[1-3]{1,2}-[1-3]{1,2}-[1-9]{4}', nullable=False),
    StringGen('[1-3]{1,2}/[1-3]{1,2}/[1-9]{4}', nullable=False),
])
@pytest.mark.parametrize('schema', [StructType([StructField('value', DateType())])])
@pytest.mark.parametrize('date_format', [
    'yyyy-MM-dd',
    'yyyy/MM/dd',
    'dd-MM-yyyy',
    'dd/MM/yyyy',
    'MM-dd-yyyy',
    'MM/dd/yyyy',
])
@pytest.mark.parametrize('ansi_enabled', [True, False])
def test_json_read_generated_dates(spark_tmp_table_factory, spark_tmp_path, data_gen, schema, date_format, ansi_enabled):

    # create test data with json strings where a subset are valid dates
    # example format: {"value":"3481-1-31"}
    path = spark_tmp_path + '/JSON_DATA'
    with_cpu_session(lambda spark: gen_df(spark, data_gen).write.json(path))

    updated_conf = copy_and_update(_enable_all_types_conf,
        {'spark.sql.ansi.enabled': ansi_enabled,
        'spark.sql.legacy.timeParserPolicy': 'CORRECTED'})

    f = read_json_df(path, schema, spark_tmp_table_factory, { 'dateFormat': date_format })
    assert_gpu_and_cpu_are_equal_collect(f, conf = updated_conf)