feat(ui): add terminal reconnection on daemon disconnect

[Wirasm]

Summary

• Add ReconnectState enum (Idle, Connecting, Failed) and daemon_session_id field to Terminal
• Add Reconnect GPUI action on TerminalView triggered by R key when terminal has error
• Reconnect handler: spawns async task calling connect_for_attach(), builds new Terminal::from_daemon(), replaces terminal + event task atomically
• Update error banner to show reconnect status ("Press R to reconnect", "Reconnecting...", "Reconnect failed: {msg}")
• Edge cases: daemon gone (clear error), session destroyed (SessionNotFound), multiple R presses (debounced), local terminals (ignored)

Test plan

• cargo fmt --check passes
• cargo clippy --all -- -D warnings passes
• cargo test --all passes
• cargo build --all passes
• Manual: open UI → create daemon session → kill daemon → press R → verify reconnection

[Wirasm]

PR Review Summary

Critical Issues (0 found)

No critical issues found.

Important Issues (2 found)

Agent	Issue	Location
code-reviewer	Reconnect uses hardcoded 24x80 dimensions. connect_for_attach(&sid, 24, 80) ignores the terminal's current dimensions. After reconnect, the daemon session starts with default size until the next prepaint triggers a resize. This causes a visible content reflow flash. Consider reading current_size from the old terminal before spawning the reconnect task and passing it through.	terminal_view.rs:243
silent-failure-hunter	reconnect_state lock failure is silently swallowed. set_reconnect_state() does if let Ok(mut s) = self.reconnect_state.lock() — if the mutex is poisoned, the state update is silently dropped with no logging. Every other lock-failure path in state.rs logs an error (see set_error_state, error_message, etc.). Add a tracing::error on the Err branch for consistency.	state.rs:834-837

Suggestions (4 found)

Agent	Suggestion	Location
type-design-analyzer	daemon_session_id could use SessionId newtype. The field is Option<String>, but the codebase has a SessionId newtype in kild-protocol. Using Option<SessionId> would prevent accidentally passing a branch name where a session ID is expected. The from_daemon constructor already receives session_id: String so this would require updating the signature too — consider if the type safety is worth the churn.	state.rs:236
code-simplifier	reconnect_state could be AtomicCell or simpler. The Arc<Mutex<ReconnectState>> is only read/written from the main thread (view layer) — never from background tasks. The Arc sharing between Terminal and the view layer is unused since set_reconnect_state/reconnect_state are only called on &self/&self through the owned view.terminal. A plain Cell<ReconnectState> or even a direct field on TerminalView (rather than on Terminal) would be simpler and avoid the mutex overhead.	state.rs:238
code-reviewer	Consider also accepting uppercase R. The reconnect keybinding only fires on lowercase r (key == "r"). If CapsLock is on, the key would be "R" and reconnect wouldn't trigger, which could confuse users. Consider key.eq_ignore_ascii_case("r") or matching both.	terminal_view.rs:336
docs-impact-agent	CLAUDE.md terminal section is not affected. The existing documentation does not reference reconnection behavior, and this is a UI-internal feature. No doc updates needed.	N/A

Strengths

• Clean separation: reconnect state lives on Terminal, reconnect logic lives on TerminalView — follows existing ownership patterns
• Proper debouncing: ReconnectState::Connecting check prevents multiple in-flight reconnects
• Old terminal cleanup is correct: swapping view.terminal drops the old Terminal, which sends Detach (best-effort) and cancels old tasks via _event_task replacement
• Good tracing coverage: _started, _completed, _failed events follow the project's structured logging convention
• Banner UX is well-considered: different messages for idle/connecting/failed states with retry hint

Verdict

READY TO MERGE (with minor fixes recommended)

Recommended Actions

1. Pass actual terminal dimensions to connect_for_attach instead of hardcoded 24x80
2. Add tracing on poisoned lock in set_reconnect_state
3. Consider suggestions at your discretion — none are blocking