fix: recover from ErrSessionMissing when remote MCP server restarts

[dgageot]

hi all - i /think/ we're observing an issue where a long-lived docker agent (via serve) fails to reconnect to MCP servers after the MCP server restarts - not recovering from errSessionMissing

Assisted-By: docker-agent

[docker-agent]

Assessment: 🟡 NEEDS ATTENTION

This PR adds reconnection logic to recover from ErrSessionMissing when MCP servers restart. The implementation is mostly sound, but there's one timing issue worth addressing.

[docker-agent]

⚠️ MEDIUM: Timeout not coordinated with retry backoff

The sessionMissingRetryTimeout is set to 30 seconds, but tryRestart() uses exponential backoff that can take up to 31 seconds total (1+2+4+8+16 seconds across 5 retry attempts).

This creates a race condition where:

1. A tool call encounters ErrSessionMissing and calls forceReconnectAndWait()
2. watchConnection is in the middle of a backoff sleep (e.g., the 16-second sleep on the 5th retry)
3. The tool call times out at 30 seconds and returns an error to the user
4. A second later, the reconnect completes successfully
5. The next identical tool call succeeds

Impact: Non-deterministic failures where tool calls fail with "timed out waiting for MCP server reconnection" even though the server successfully reconnects moments later.

Recommendation: Either increase sessionMissingRetryTimeout to 35-40 seconds to account for the maximum backoff duration, or coordinate the timeout with the actual retry logic (e.g., calculate remaining backoff time).