[tintinhamans]

Merge after #656. Conflicts with #625 in NetworkMesh.h and NextGenTransport.cpp; this stack gets rebased once #625 lands.

There were no numbers for how a connection behaves beyond the lobby ping, and the only loss figure was a guess from command retries. ConnectionStats (new) keeps latency, jitter and loss of one connection over a 10 s window, and a retry whose earlier send was acked after all counts as a late ack, not as loss. The mesh also probes each peer on the ping channel, every connection.probe_lobby_interval_ms in the lobby (250) and connection.probe_match_interval_ms in a match (1000), timed from when the transport received the probe so a slow frame is not counted as network delay.

PacketFeedback (new) measures real packet loss per direction. Every game packet gets a sequence number and the receiver reports what it got over the last 64 packets, at most every 250 ms. It only sees byte buffers and a clock, so GNS, EOS or any other transport can share it.

Wire change: game packets start with a 3 byte envelope (u16 sequence, u8 flags) plus a 5 byte feedback record when due, and GENERALS_MAGIC_NUMBER is 0xF00E, so this build does not talk to an older one. A packet without a valid envelope, or on an unknown channel, is dropped. A failed GNS send is retried once and returned as a failure. The send.flags key is removed. The numbers show in the debug stats display.

Tested: builds x86 Release z_generals. The full stack had 0 CRC mismatches in 3 player rig runs, n=1.