Implement async IPC

[0forks]

IO is now handled by libuv. Pose requests now happen at 400-500hz to minimize latency, decoupled from RunFrame.

• Added some basic tests;
• Dependencies are now pinned using vcpkg submodule as recommended, readme and CI updated accordingly;
• Reduced binary size (stripped debug symbols on linux, switched to lite protobuf runtime);
• Fixed some code style inconsistencies.

Tested with ALVR, WMR on Windows combined with SlimeVR/SlimeVR-Server#629.
Reported by other testers that works on Linux.
Needs more testing on Linux.

Beta checklist

OS:

• Linux
• Windows

HMD driver:

• Native SteamVR / Lighthouse tracking (Vive, Pimax, ...)
• Virtual Desktop
• Steam Link
• Meta Air Link (can skip)
• ALVR
• WMR
• Pico Connect (can skip)

[kitlith]

Sorry it's taken so long for me to review this. Overall, I think you've raised the code quality of the driver, even if I wish some of the style/formatting changes had been done in a separate PR to make this one easier to review. (However, I never spoke up previously to comment on or ask for that, so that's on me.) Almost every time I went "wait, that's weird" it ended up being original code that was reformatted, or a quirk of the original code that was faithfully reproduced, and I'm not going to argue with that.

None of my review comments block this PR. (but to be clear, I would appreciate some response.) They fall in a few categories:

• Questions that I'd like an answer for, but require no action and I won't loose sleep over not knowing.
• nits about byteswap that I'd prefer were addressed, but are ultimately unimportant. (it'll work either way.)
• pointing out things that others might find interesting.

There are a couple limitations to what I've reviewed at this time:

• I haven't viewed libuv documentation, so I'm basically going off of vibes, and taking it on faith since there are multiple reports of this code working.
• Similarly, I noticed that you've started using atomics in some places to account for the newly multi-threaded nature of this code. I don't have the spoons to check if you've missed anything. (I haven't done any multithreaded programming in C++, and Rust will largely just fail to compile if you haven't justified yourself correctly.)
• I've mostly glossed over the test code.

[0forks]

Tracker reconnection/readding broke after reverting pose = MakeDefaultPose(); 208d86e, now we explcitly set deviceIsConnected/poseIsValid flags when updating pose 279b469.

Originally on each reconnection there was a new device instance created with these flags set to true by default, now we're reusing the existing devices.