Add Cate Agent mission orchestration

[Anton-Horn]

What changed

• add a Cate Agent orchestration extension that can launch, supervise, wait for, and stop multiple coding agents
• surface agent activity and orchestration tool results in the chat UI with structured status cards
• scope worktree selection to chats so the selected target follows the chat across sidebar and panel hosts
• seed agent panels created from Parallel Work with a worktree-bound chat and clean up legacy panel worktree residue
• improve composer popover placement, empty chat states, and terminal stop behavior while preserving scrollback
• include the repository-local Cate CLI and Karpathy guideline skills
• keep healthy coding-agent terminals in starting/working instead of treating the registry's no-failure null as a failure
• return actionable CLI exit diagnostics and teach the supervisor to retry CLI-specific failures with another registered agent
• avoid a redundant worker-creation confirmation after the user explicitly enables orchestration mode

Why

Cate Agent needed an explicit orchestration path for coordinating coding agents without conflating a panel host with the worktree a chat operates in. Keeping the worktree on the chat makes ownership and persistence consistent when chats move between hosts, while the orchestration UI makes delegated work observable and controllable.

The terminal failure registry returns null when there is no startup failure. Mission status consumers compared that value with undefined, so every newly created Codex worker could be reported as failed even while its PTY was alive and working.

User impact

Users can ask Cate Agent to coordinate parallel coding work, see each delegated agent's real progress, receive useful failure reasons, retry provider-specific failures with another CLI, stop running work cleanly, and keep each chat attached to the correct worktree as it moves around the app.

Validation

• npm run typecheck — passed after merging current main
• focused post-merge Vitest coverage for the orchestrator and coding-agent status paths — 4 files / 18 tests passed
• full npm test host run — 3057 passed, 45 skipped, 1 unrelated failure in src/main/runtime/local-daemon-tarball.test.ts; its temporary workspace was rejected by the runtime allowlist (Access denied: path ... is outside allowed directories)