Add validation to reader benchmarks

[vuule]

Description

Validate that cuIO readers successfully round trip the input tables.
Validation is only done in the first iteration, and it is not included in the timing.
If there is a difference, a warning is logged.
Setting CUDF_BENCH_OUTPUT_DIFF environment variable adds diff to the standard output. Valid values are FIRST_ERROR and ALL_ERRORS.

Benchmarks currently report differences for data types that can't be preserved through the given formats. There are not differences caused by data corruption.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vyasr]

The implementation seems sensible, but why are we doing this in benchmarks rather than tests?

[vuule]

The implementation seems sensible, but why are we doing this in benchmarks rather than tests?

To verify that the benchmarks are doing what they're supposed to. Also, it's helpful in situations where we benchmark different configurations (thread pool size, GDS policy, compression policy...). Basically enabling us to benchmark changes without jumping back to tests for each change.