Increase UDP socket receive buffer to prevent large message loss

[alvgaona]

Problem

RustDDS silently drops large DDS messages (e.g., sensor_msgs/Image at ~1200p or above) because the OS-default UDP receive buffer is too small to absorb fragment bursts. When a publisher (e.g., CycloneDDS) sends a large message, it gets fragmented into thousands of small UDP packets (~1 KB each). These arrive in a burst that overflows the default receive buffer (~786 KB on macOS, ~208 KB on Linux). With BestEffort QoS — standard for image topics — lost fragments are never retransmitted, so the message is silently lost.

Fix

Set SO_RCVBUF to 8 MB on all UDP listener sockets. The buffer size is configurable via DomainParticipantBuilder::socket_receive_buffer_size().

On macOS, the per-socket buffer is capped at kern.ipc.maxsockbuf (8 MB by default), which limits reliable reception to ~8 MB payloads. On Linux, users can raise net.core.rmem_max for larger payloads (e.g., 4K uncompressed images).

Testing

Tested with a ROS 2 camera node publishing sensor_msgs/Image over CycloneDDS:

Resolution	Payload	Before	After
320×240	230 KB	✓	✓
640×480	920 KB	✓	✓
1100×540	1.8 MB	✓	✓
1200×540	1.9 MB	✗	✓
1920×1080	6.2 MB	✗	✓
3840×2160	24.9 MB	✗	✗ (macOS buffer cap)

For high-resolution images those must be compressed. Apparently in Linux the max_buf_size is more generous than macOS and would allow raw 4K streams.

[alvgaona]

@jhelovuo more enhancements. Let me know what you think.