Telemetry: instrument connection lifecycle (SSH process, WebSocket)

[EhabY]

Part of the Telemetry Phase A rollout. See the RFC in Linear: AIGOV-154.

Events

• ssh.process.discovered with result, measurement durationMs.
• ssh.network.info with p2p, derp, measurements latencyMs, downloadMbits, uploadMbits. Sampled at 60 seconds or on a meaningful change (p2p flip, derp flip, or >10% latency swing), whichever comes first. Decoupled from the 3-second networkPollInterval in sshProcess.ts.
• ssh.process.lost with cause, measurement uptimeMs.
• ssh.process.recovered with measurement recoveryDurationMs.
• connection.open with url, measurement connectDurationMs.
• connection.drop with cause, closeCode, measurement connectionDurationMs. On unexpected close, include an error block.
• connection.reconnect with result, measurements attempts, totalDurationMs.
• connection.state_transition with from, to, reason. One event per transition in the 6-state machine: IDLE, CONNECTING, CONNECTED, AWAITING_RETRY, DISCONNECTED, DISPOSED.

Sites

• src/remote/sshProcess.ts discovery, loss, and recovery paths. Add the network-info sampler that reads the existing polled file but emits on the 60s-or-on-change cadence.
• src/websocket/reconnectingWebSocket.ts state reducer and handlers.

Tests

• Per-event assertions via TestSink.
• Network info sampling emits on first read, after 60 seconds, and on a p2p flip inside the 60s window. Does not emit per 3-second poll.
• State transition events fire on every reducer change.
• Reconnect cycle aggregates attempts and totalDurationMs correctly.

Depends on AIGOV-243.