feat: Add Hugging Face dataset benchmarking support

- Implement HuggingFaceDatasetLoader for easy dataset loading and preprocessing
- Add BenchmarkRunner with comprehensive performance metrics (latency percentiles, throughput)
- Integrate dataset benchmarking into CLI with --dataset flag
- Support multiple output formats (JSON, CSV) for benchmark results
- Include example scripts demonstrating benchmarking workflows
- Add tests for new benchmarking functionality
- Update documentation with benchmarking usage examples

This enhancement allows users to easily benchmark their deployed models
on real-world datasets from Hugging Face Hub, providing detailed
performance insights across different batch sizes.