🧠 Substantial refactor of connection and routing logic

[haydenmc]

This payload is a substantial refactor of several parts of the project to address longstanding technical debt and pave the way for automated testing.

• Network transports have been abstracted into their own interfaces and implementations, allowing for re-use and trivial mocking for classes that depend on transmitting and receiving network data.
• FTL server implementation has been decoupled from JanusFtl and refactored into several classes with clearer separation of responsibilities: FtlServer, FtlControlConnection, and FtlStream.
• FtlStream class now works to maintain the order of incoming RTP packets, more intelligently sends NACKs for groups of missing packets instead of one-at-a-time, and calculates additional stats such as running average bitrate.
• Fewer copies of RTP packets are made as they are received and transmitted to viewers.
• Only one stream metadata update thread runs instead of one for each stream.
• Clearer locking patterns (and shared/exclusive locks instead of lock guards) for enforcing thread safety.
• Simplified JanusFtl handling of FtlStream objects and associated JanusSession viewers, removing FtlStreamStore.
• Removed RelayThreadPool, as the performance benefits were not yet made clear, and it often resulted in packets being delivered out-of-order
• Ownership patterns made more clear through replacement of many shared_ptr instances with unique_ptr and plain reference passing.
• Removed dependency on Janus helper functions JANUS_LOG and rtp.h utility functions.
• Result<T> type introduced to better indicate success/failure of an operation.
• Introduced Catch2 for automated testing

[danstiner]

Been testing this branch, it's definitely more reliable and at 5mbps I see somewhat decreased packet loss and grey frames.

Using callgrind I see no obvious places in code that waste lots of CPU, previously thumbnail generation alone was ~50% of CPU usage and relay threads another ~25% (seemingly due to constructing a byte vector in an inefficient way)

@haydenmc why not just merge this as is and then add back thumbnail generation? I mean it works great :)