The main benefit of libyuv, apart from shortening the video pipeline, is quite noticeable latency and CPU usage decrease due to various assembler/SIMD optimizations of the library. However, there is a drawback for macOS systems: libyuv cannot be downloaded as a compiled library and can only be built from the source, which means we should include a cropped source code of the library (~10K LoC) into the app or rise the complexity of macOS dev and run toolchains. The main target system -- Linux, and Windows will use compiled lib from the package managers and macOS will use the lib included as a shortened source-code.
Building the app with the no_libyuv tag will force it to use libyuv from the provided source files.
In the current version of the application, we have strictly hardcoded the captured runtime application (FFI Libretro frontend) as well as the streaming transport (WebRTC). This commit makes it possible to choose these components at runtime.
In this commit, we no longer manage initially connected users separately from the rooms, and instead, we treat all users as abstract app sessions, rather than hardcoded WebRTC connections. These sessions may contain all the transport specifics, such as WebRTC and so on.
Rooms, instead of having the hardcoded emulator app and WebRTC media encoders, now have these components decoupled. In theory, it is possible to add new transports (e.g., WebTransport) and streaming apps (e.g., wrapped into an ffmpeg desktop app).
