Add native RDP server backend

[kogekiplay]

Summary

This draft PR adds a native RDP server backend for ReFrame, behind an optional Meson feature flag. The goal is to let ReFrame expose the existing capture/conversion/input/session pipeline over RDP, with FreeRDP used as a reference for the protocol implementation.

This is intentionally opened as a draft because the branch is large and the protocol surface is broad. I would like early feedback on the direction, module boundaries, and which pieces should be reduced or split before review.

Implemented

• Adds a generic RfRemoteServer abstraction so the main server pipeline can use VNC or RDP backends through the same start/stop/update/flush/input/resize/clipboard-style interface.
• Keeps existing VNC behavior as separate libvncserver/neatvnc modules.
• Adds [remote] protocol=vnc|rdp configuration, with VNC remaining the default for compatibility.
• Adds optional RDP build support through Meson, so default non-RDP builds do not require the RDP codec/tooling stack.
• Implements the native RDP connection path: TPKT/X.224, TLS, MCS/GCC, security exchange, licensing-valid response, Demand Active/Confirm Active, client finalization, input PDUs, and update dispatch.
• Implements RDPGFX dynamic virtual channel handling, including capability negotiation, surface creation, frame lifecycle, WireToSurface, FrameAck/QoE feedback, fragmentation, and ZGFX payload compression.
• Implements graphics update paths for AVC420, AVC444/AVC444v2, AV1 i420/i444, Progressive, RemoteFX, PLANAR, NSCodec, and bitmap fallback.
• Adds codec quality/backpressure policy for bitrate, QP, GOP, keyframe cadence, AVC444 LC/delta behavior, and QoE/FrameAck-based adaptation.
• Adds keyboard, pointer, wheel, resize ownership, and single-active-client replacement policy.
• Implements cliprdr text, HTML, and common image clipboard paths, including CF_UNICODETEXT, HTML Format, DIB/DIBV5, BMP, PNG, JPEG, WEBP, and TIFF handling.
• Implements RDP audio output over rdpsnd, including static channel negotiation, PCM and IMA/DVI ADPCM format selection, Wave/Wave2 PDUs, WaveConfirm/training/quality-mode handling, PipeWire monitor capture, configurable sample rate/channels/frame size, sender queueing, latency statistics, silence suppression for better A/V sync, and audio-codec=auto|pcm|adpcm configuration.
• Adds unit tests for protocol parsing/writing, MCS, DVC, RDPGFX, input mapping, config parsing, resize policy, AV1/AVC helpers, RemoteFX, NSCodec, cliprdr, clipboard rich-format helpers, RDP audio stream/rdpsnd helpers, ADPCM format/encoding helpers, and core update/quality policies.

Not Implemented / Future Work

• NLA/CredSSP/NTLMv2 authentication is not implemented yet; TLS is supported.
• Dynamic Display Control / monitor-layout updates are not complete yet; the current backend handles initial client desktop size and resize ownership policy.
• Full cursor shape support is incomplete; the backend has a basic/default pointer path but not full Color Pointer / Large Pointer handling.
• File clipboard support is not implemented yet, including CF_HDROP, FileGroupDescriptor, FileContents, drag/drop-style file streams, and related cancellation/error paths.
• More complete delayed clipboard rendering and format-on-demand behavior is future work.
• More exhaustive Windows/FreeRDP clipboard format compatibility, edge formats, and application-private format passthrough are future work.
• Very large clipboard payload flow control, cancellation, and recovery behavior can be improved further.
• Additional rdpsnd encoded formats beyond PCM and IMA/DVI ADPCM are future work.
• Advanced audio device controls, volume synchronization, and richer client audio policy handling are future work.
• Multimedia redirection/TSMF is not implemented.
• Drive redirection, printer redirection, smartcard, serial/parallel redirection, audio input redirection, and RemoteApp/RAIL are out of scope for this PR.

Audio Scope

The audio path implemented in this PR covers the main remote-desktop playback case: server-side audio capture from PipeWire and client playback through rdpsnd, with low-latency scheduling, confirmation handling, and PCM/IMA ADPCM output depending on client support. Audio input redirection is intentionally out of scope for this PR because it is more relevant to conferencing or microphone-forwarding workflows than to the core cross-platform remote desktop path.

Test Plan

Automated checks run on the RDP development target:

• meson test -C build-rdp --print-errorlogs
• Targeted checks during the latest audio work:
  • meson test -C build-rdp reframe:rdp-rdpsnd reframe:rdp-audio-stream reframe:rdp-core reframe:config --print-errorlogs
• meson setup build-nordp -Drdp=false
• meson compile -C build-nordp
  • non-RDP build still compiles without the RDP backend enabled

Manual smoke testing performed during development:

• Connected with Remote Desktop Manager and compatible RDP clients.
• Verified RDPGFX Progressive fallback with RDM.
• Verified AVC444/AVC420 and AV1-oriented paths with compatible RDP client options.
• Verified that only one active RDP client is kept; a later connection replaces the previous one to avoid mixed-resolution/QoE/input conflicts in the current single-desktop architecture.
• Verified text, HTML fallback, and common image clipboard transfer in both directions with compatible clients.
• Verified live rdpsnd audio output with a compatible client, including Wave2 playback, confirmation handling, silence-suppression behavior for improved A/V sync, and compressed ADPCM negotiation fallback through audio-codec=auto.
• Deployed to a test prefix and exercised live capture/update/input/audio paths against a running ReFrame session.

Notes for Reviewers

• The PR is large because it contains the protocol backend, codec matrix, clipboard support, and audio output path. I am happy to split it if that makes review easier.
• The RDP backend is meant to be optional and should not change existing VNC defaults.
• The current design treats RDP as a new protocol backend rather than fitting it under the existing VNC naming/interfaces.

[AlynxZhou]

I think there is no need to add another section, this option could be part of [reframe] group.

[AlynxZhou]

There is no reason that we need rpath.

[AlynxZhou]

At least currently we have no way to get audio via streamer, what is this handling?

[AlynxZhou]

Are you coding with AI/LLM? Check your produced output, I don't think things have to be done in such a complex way, your PR smells like pouring code anywhere, only crazy man will allow such kind of code to be merged. Check your code, don't waste my time to review/refactor your LLM code, YOU should do it.

[AlynxZhou]

I don't think we need to handle rich format clipboard, I am even not sure whether VNC supports it, so don't do this, at least for this PR, focus on core RDP.