Harden multiplayer networking and adapt frame timing

[ZivDero]

Reworks thbe networking:

1. Hardens network packet handling
   Defended against too small, too large packets, malformed packets, unknown packets.
   Also covers the entire abuse set of sending packets as another player's house, so now the only way to cheat here is to fake stuff on the actual network transport layer.

2. Improve connection-level timing
   Partly based on Add fast retransmit of lost packets and packet redundancy CnCNet/yrpp-spawner#75
   Improves actual connection timing to better reflect connection quality on a per-connection basis to not stall everyone unnecessarily.

3. Improve lockstep timing
   An evolution of "protocol zero", improving the selection of FrameSendRate and MaxAhead based on the connection quality.
   Includes downgrading for bad connections and limited upgrading (to prevent ping-pong when connection quality varies).

Closes #21, closes #22, closes #23, closes #24, closes #25, closes #26, closes #35

[github-actions]

Development builds of 4230f5a:

• opents-Release-c66fff8
• opents-Debug-c66fff8

The links work without a GitHub account. Artifacts expire after 90 days, and this comment follows the latest successful build.

[ZivDero]

Gonna split this up

[11EJDE11]

Nice. Stacked PRs might help? https://github.github.com/gh-stack/

I tested it earlier and it was a worse experience. Added some logging and fed them to Codex:

So my conclusion is: the PR’s adaptive approach can work, but its current RTT estimator has a sampling-starvation bug
under a sudden latency increase. It then chooses exactly the wrong timing—too little headroom and too much packet
traffic. The next fix should age or invalidate RTT samples when no eligible acknowledgements arrive, carry
retransmission backoff into subsequent packets, and make repeated retransmissions force conservative timing rather
than continuing to report a fresh 10 ms RTT.
networkstats.zip

Also there were issues getting mpdebug working. I haven't checked or verified, but Codex said:

• It’s legacy design, not a LAN issue. MPDEBUG was originally a reserved panel rather than a true overlay:

• Physical output: 640×480.
• Gameplay viewport: 640×400.
• Diagnostics: rows 400–479.

The bgfx renderer migration accidentally collapsed those dimensions. C:/Users/QWE/source/repos/OpenTS/code/
startup.cpp:605 now sets the entire video surface to 640×400 before creating it, while C:/Users/QWE/source/repos/
OpenTS/code/mainloop.cpp:841 draws from y=400 onward. Everything is therefore clipped outside valid rows 0–399.

Simply removing the resolution override is insufficient: Multiplayer_Debug_Print() currently runs before Map.Render(),
whose final blit would overwrite the diagnostics.

[ZivDero]

MPDEBUG is finnicky in TS yeah. In Vinifera I implemented an ImGui window that included it. Maybe now is a good time for that.