fix(remote): make reconnect resilient

[abeljim]

Summary

• keep failed remote bridge panes visible with an actionable error instead of closing into an empty layout that resurrects at local root
• resize retained remote PTYs before replay and skip incomplete VT history when the replay buffer was truncated
• close completed remote PTYs before rerun while preserving intentional clean-exit behavior
• add a flock-sidebar message that switches to the next visible session before killing the current session
• make session cycling follow sidebar display order, wrap at the end, and safely do nothing when no alternative session exists
• document the optimized local development and reconnect workflow in AGENTS.md

Testing

• cargo fmt --all -- --check
• cargo test remote_agent (29 passed)
• cargo test -p zellij-server remote_pty
• cargo test -p zellij-server remote_shell_exit
• cargo test -p zellij-server direct_remote_command
• cargo test -p zellij-server ordinary_held_commands
• cargo test -p flock-sidebar (306 passed)
• cargo xtask build --release --plugins-only
• manually reproduced and verified Coder session reconnection, including a replay buffer filled with repeated devenv status redraws