[FEA] Enhance integration test to avoid test failures with different seeds

[thirtiseven]

Is your feature request related to a problem? Please describe.

Currently, the random seed in integration tests is set to 0 by default, and all cases can pass with this seed. However, if we change the seed to other values, some cases will fail.

I wish we could improve the integration tests to avoid this, e.g. use a dynamic seed for each run, or test with more data for those cases that are easy to fail.

Some failed cases:

• test_cast_fallback_not_UTC failed with seed 4 #9242
• test_groupby_first_last failed with seed 4 #9243
• test_str_to_map_expr_fixed_pattern_input failed with seed 4 #9244
• test_hash_groupby_collect_list failed with seed 4 #9245
• [BUG] test_predefined_character_classes failed with seed 4 #9246
• test_initcap failed with seed 14 #9247

We should fix them or mark them as XFAIL first.

Describe the solution you'd like
Set random seeds for each test run and log the seed for reproducibility.

Describe alternatives you've considered
Or we can make the easy to fail cases stronger, e.g. more data for aggregate tests and tests with long generated strings, more precise data range for fallback/exception tests.

Additional context
Some similar issues that has been fixed: #8716 and #8607

[revans2]

The init_cap issue is something that CUDF says that they will fix. All of the others have had fixes merged in.