fix(board): preserve sessions when panel restore fails

[peters]

Summary

Fixes #174 by making runtime-state restoration tolerant of individual panel spawn failures.

When one panel cannot be restored, Horizon now:

• keeps restoring the remaining panels in the session
• creates a terminal-backed placeholder for the failed panel with the restore error
• preserves the failed panel original kind, command, args, cwd, layout, local id, and session metadata so it can be restarted after the underlying problem is fixed
• adds a high-severity attention item for the failed restore instead of silently falling back to an empty board

Root Cause

Board::from_runtime_state_with_transcripts() used ? while restoring each panel, so any single create_panel() failure aborted the whole board restore. The UI then caught that board-level error and replaced the session with Board::new(), making every panel appear lost even though runtime.yaml still existed.

Validation

• cargo fmt --all -- --check
• ./scripts/check-maintainability.sh
• cargo test --workspace
• cargo clippy --all-targets --all-features -- -D warnings

Note: while investigating corrupt or missing Codex binaries, I found that some executable failures can surface as fast child exits instead of parent-side PTY spawn errors on macOS. This PR fixes the board-abort path described in the issue: any restore-time create_panel() error no longer takes down the full session.