Add comprehensive benchmarks for MCAP library

[siliconlad]

This commit adds a comprehensive benchmark suite comparing pybag against rosbags and official mcap-ros2 libraries across multiple dimensions:

Serialization benchmarks:

• Odometry, IMU, LaserScan, PointCloud2, String message types
• Mixed message type serialization performance

Deserialization benchmarks:

• All message types above with pybag, rosbags, and mcap-ros2
• Measures decoding performance from CDR format

Read benchmarks (extended):

• Small MCAP (10 messages) for startup overhead analysis
• Multi-topic MCAPs with 6 different message types
• Multi-topic with images/point clouds (7 topics with large data)
• Extended duration (10 second) multi-topic scenarios

Write benchmarks (extended):

• Small (10 messages), standard (1000), and large (10000) message counts
• Multi-topic writing with interleaved timestamps

HuggingFace dataset integration:

• Download utility for real-world MCAP files from DapengFeng/MCAP
• Benchmarks against diverse, production ROS2 bag files

Multi-topic MCAP generation utilities:

• Generators for Odometry, IMU, LaserScan, PointCloud2, Image, etc.
• Configurable topic frequencies and recording duration