Refactor benchmarking scripts

https://github.com/NVIDIA-NeMo/Curator/pull/1382 helped clean up some repeated logic by creating a utils.py script with helper functions.

While working on that, I noticed other areas of improvement. For example, we pass the same arguments a lot of the time, so we could consider creating a helper function for it:

def add_common_benchmark_args(parser: argparse.ArgumentParser) -> None:
    """Add common benchmark arguments to an argument parser.

    This adds the standard arguments used across most benchmarks:
    - benchmark-results-path (required)
    - input-path (required)
    - output-path (optional, with default)
    - executor (optional, with choices)

    Args:
        parser: ArgumentParser to add arguments to
    """
    parser.add_argument("--benchmark-results-path", type=Path, required=True, help="Path to benchmark results")
    parser.add_argument("--input-path", required=True, type=Path, help="Path to input data")
    parser.add_argument("--output-path", type=Path, help="Output directory for results")
    parser.add_argument("--executor", default="ray_data", choices=["xenna", "ray_data", "ray_actors"], help="Executor to use")

A lot of the success/failure logic is redundant between scripts too. Helper functions like:

def run_benchmark_main(
    benchmark_func: Callable[[argparse.Namespace], dict[str, Any]],
    args: argparse.Namespace,
    benchmark_name: str,
) -> int:
    """Standard main function wrapper for benchmarks.

    Handles the try-except-finally pattern and result writing that's common
    across all benchmark scripts.

    Args:
        benchmark_func: Function that runs the benchmark and returns results dict
        args: Parsed command-line arguments
        benchmark_name: Name of the benchmark for logging

    Returns:
        Exit code (0 for success, 1 for failure)
    """
    logger.info(f"=== {benchmark_name} Benchmark Starting ===")
    logger.info(f"Arguments: {vars(args)}")

    try:
        results = benchmark_func(args)

    except Exception as e:  # noqa: BLE001
        error_traceback = traceback.format_exc()
        print(f"Benchmark failed: {e}")
        logger.debug(f"Full traceback:\n{error_traceback}")
        results = {
            "params": vars(args),
            "metrics": {
                "is_success": False,
            },
            "tasks": [],
        }

    finally:
        write_benchmark_results(results, args.benchmark_results_path)

    # Return proper exit code based on success
    return 0 if results["metrics"]["is_success"] else 1

could help us create a cleaner experience.