SITL support for UDP connected network peripherals

[timtuxworth]

Summary

Adds UDP support for simulated network-attached devices (--net-device), alongside the existing TCP-only path.

Classification & Testing (check all that apply and add your own)

• Checked by a human programmer
• Non-functional change
• No-binary change
• Infrastructure change (e.g. unit tests, helper scripts)
• Automated test(s) verify changes (e.g. unit test, autotest)
• Tested manually, description below (e.g. SITL)
• Tested on hardware
• Logs attached
• Logs available on request

Built SITL (./waf copter) with these two commits in isolation and confirmed the existing MountTopotekNetwork autotest (TCP path, unaffected by this change) still passes.

Description

SerialDevice::listen_on_tcp_port() was the only way to attach a simulated network-attached device (--net-device NAME:PORT, used for devices reached over a NET_Pn port rather than a serial port). Some real hardware being simulated is UDP-only, so this adds:

• SerialDevice::listen_on_udp_port() and a network_update_udp() counterpart to network_update(). Unlike TCP there's no connection to accept — the autopilot's address is learned from whichever datagram it last sent, and replies go back to that address.
• create_net_serial_sim() now accepts an optional protocol after the port, e.g. topotek:15005,udp (TCP remains the default with no suffix). The spec format changes from NAME:PORT to NAME:PORT[,PROTOCOL] — the protocol is comma-grouped with the port rather than colon-chained, since it (and any future per-port option, e.g. a simulated firmware version) is logically attached to the port, not another top-level field like NAME.
• an autotest for Topotek gimbals over UDP

Split out of #34155 (a new SkyDroid gimbal driver, whose real hardware is UDP-only) at Peter Barker's suggestion that this capability doesn't depend on anything SkyDroid-specific and is useful on its own. That PR will rebase on top of this once merged.

---

This PR was developed with AI assistance (Claude). All changes have been reviewed and tested by me as part of #34155 .

[tridge]

Automated review note — AI-generated (Claude), validated against the live diff. Please sanity-check before acting.

Full report: https://uav.tridgell.net/DevCallReviews/AIReview/devcall_pr_reviews.html#pr34159

Re-reviewed at head 0ef30f5d6f: all three substantive findings from my previous review are resolved, and the UDP path is now exercised end-to-end in CI (sitltest-copter-tests2b runs MountTopotekNetworkUDP and passed). Verdict: COMMENT — mergeable after cosmetics.

• UDP >300-byte truncation — resolved (SIM_SerialDevice.cpp:287 caps at MIN(sizeof(buffer), 300) with a comment explaining the coupling; MIN(300U, space) re-verified on master's AP_Networking_port.cpp:332).
• "is dropped" comment — resolved (now accurately says pre-peer data is left unread, nothing lost).
• Stale "via TCP" sim_update() comment — resolved.
• Stale PR body — still open: it still says "No new autotest is added here" and "these two commits"; the PR has 5 commits including the new autotest.

Remaining, none blocking:

1. SIM_SerialDevice.cpp:181 — SO_REUSEADDR on the UDP listen socket lets a second SITL instance bind the same port "successfully" with datagrams silently going to one socket, instead of a clean EADDRINUSE panic. Note dropping the explicit reuseaddress() call isn't enough — SocketAPM::bind() itself calls it unconditionally (Socket.cpp:250) — so this needs a bind-API variant or a deliberate decision to accept it.
2. SITL_State_common.h:100 — the declaration comment still documents TCP-only NAME:TCPPORT; the spec is now NAME:PORT[,PROTOCOL] (and, trivially, a third comma token like name:port,udp,garbage is silently accepted).
3. Commit tidy-up before merge: 49a9971 and 0ef30f5 are amendments to earlier commits in this PR; squashing them into their base SITL / AP_HAL_SITL commits would keep the clean 3-commit structure (preference, not a gate — convention CI passes as-is).
4. Coordination: Skydroid driver for C11 and C13 gimbals #34155 still carries an older copy of this transport without the 300-byte cap — whichever lands first, the other needs a rebase.

The stm32h7 build failure is an unrelated Micro-XRCE-DDS-Gen install flake (fails in Gradle before any compilation) — worth a re-run for a fully green board. Codex cross-checked: 6 confirmed, 1 adjusted (the bind() point in item 1), 2 added (item 2).

[timtuxworth]

FYI @tridge