feat(remote): exponential reconnect backoff + classified disconnect reason

[brettchien]

Supersedes #58 (auto-closed by GitHub when its stacked base branch fix/remote-session-graceful-teardown was deleted on #57's merge). Same change, now rebased onto main (base = main).

Studio-review item 2 (reconnect backoff + error classification), 斷線原因可觀測性, and the testable slice of item 3.

What

The reconnect loop retried on a flat 5s cadence forever and surfaced every failure as a raw error string. With the gateway's small session-slot pool (max_sessions), a flapping link hammered it on a fixed beat, and the operator couldn't tell a network blip from an auth reject or the server being at capacity.

Changes

• acp-tunnel — new pure reconnect module (unit-tested; compiles/tests without GTK so the workspace CI gate covers it):
  • backoff_delay(attempt, salt) — exponential 1→2→4→8→16s, capped 30s, + per-connection jitter.
  • DisconnectReason::classify(&str) + label() — network / auth rejected / server at capacity / protocol / other.
• remote.rs run_reconnecting — escalate backoff on consecutive fast failures; reset once a connection held ≥15s; put the classified reason in the status line (error: server at capacity, …) + Activity log. Console renders the status string as-is, so the reason surfaces with no frontend change.

Tests / verification

• 3 new unit tests (classify the real DNS-storm string; backoff shape/cap; jitter bound) — green locally (cargo test -p acp-tunnel). Session state-machine already covered in acp-tunnel.
• remote.rs compiles under bundle-macos CI (no GTK on the authoring box).

Not in this PR

• Sidecar auto-restart (mcp.rs has no respawn on core crash) — separate PR.

🤖 Generated with Claude Code