uxrce_dds_client: avoid stalls under high UDP load

[nanwanuser]

• Avoid double-close of transport fd on deinit
• Reduce loop latency and drain inbound bursts
• Batch best-effort output flush and handle buffer-full
• Lower default uORB->DDS poll interval to allow higher /fmu/out rates

Solved Problem

Relates to #26160

High-rate ROS 2 -> PX4 traffic over Ethernet/UDP (uXRCE-DDS) can cause uxrce_dds_client to stall:

• /fmu/out/* topics stop publishing
• uxrce_dds_client status stops updating (counters frozen)
• PX4 stops sending UDP data and does not recover after stopping ROS 2 publishers (reboot required in my tests)

This PR is primarily intended to surface the issue upstream and contribute one mitigation/fix approach I tested on PX4 v1.16.0.

Solution

• Avoid potential double-close of the transport fd in deinit() (let the transport close its own fd)
• Configure UDP socket as non-blocking (FIONBIO) to prevent blocking send() under heavy inbound load
• Reduce loop latency (short uORB poll timeout; avoid blocking when transport has pending data)
• Drain inbound bursts by running uxr_run_session_timeout() multiple times per loop while data is pending
• In DDS topic bridge, batch best-effort flush and retry once on UXR_INVALID_REQUEST_ID to handle output buffer full
• Lower default uORB->DDS poll interval to 4 ms (~250 Hz) to raise the default /fmu/out/* forwarding rate ceiling.
  Rationale: the system already supports high-rate /fmu/out/* publishing, but the previous default poll/forwarding interval (10 ms) effectively capped the default forwarding rate; reducing it increases the maximum forwarding rate available out of the box and helps keep /fmu/out/* rates stable under high
  inbound /fmu/in/* load.

Changelog Entry

For release notes:

Bugfix: uxrce_dds_client: avoid stalls under high UDP load and improve throughput under bursts

Alternatives

• Reduce ROS 2 publish rate / add throttling on application level
• Increase XRCE stream buffer sizes (if feasible on target)

Test coverage

• PX4 v1.16.0: build succeeded and the stall was mitigated in my setup.
• main: make px4_fmu-v6x (note: current main hits FLASH overflow in my environment; this change adds ~136 bytes in .text).
• Bench test:
  • board: px4_fmu-v6x
  • ROS 2: Humble
  • publish set: /fmu/in/actuator_motors
  • rate: 1200 Hz
  • result: no stall and /fmu/out/* rates stable

Context

• Hardware: Pixhawk v6x
• Reproduction steps: see uxrce_dds_client stalls under high-rate ROS 2 UDP traffic: /fmu/out stops and status freezes #26160