[BUG] test_cast_string_ts_valid_format fails with DATAGEN_SEED=1699978422

[abellina]

Repro:

SPARK_RAPIDS_TEST_DATAGEN_SEED=1699978422 ./run_pyspark_from_build.sh -k test_cast_string_ts_valid_format

FAILED ../../../../integration_tests/src/main/python/cast_test.py::test_cast_string_ts_valid_format[String2][DATAGEN_SEED=1699978422, INJECT_OOM] - AssertionError: GPU and CPU timestamp values are different at [1614, 'a']

[jlowe]

Looked into this a bit, and I think there's two problems here. First, the test generates strings that almost always are invalid timestamps, so the test is not very useful in practice. However with this specific datagen seed, it happens to generate a valid timestamp in the third row, specifically 7141-09-13 08:15:02+121024 which parses to 7141-09-12 15:04:38 on the CPU but parses to null on the GPU.

[thirtiseven]

In case of failure: StringGen('[0-9]{1,4}-[0-3][0-9]-[0-5][0-9][ |T][0-3][0-9]:[0-6][0-9]:[0-6][0-9].[0-9]{0,6}Z?')

The . here is meant to be a literal ., but it's a wildcard character in regex, so it can generate anything in this case, which also leads to many invalid timestamps.

In the failed case it generated '+': 7141-09-13 08:15:02+121024 and Spark somehow supports this format.