feat: skip service restarts in grove tui when services already running

[windoliver]

Problem

Every grove tui invocation restarts the HTTP server and MCP server even when they are already running. This makes re-attaching to a session feel slow — especially noticeable when switching tmux windows or after a TUI crash.

Expected behavior

grove tui should detect already-running services and skip their startup, connecting directly to the existing Nexus backend and proceeding to the session picker immediately.

Proposed fix

In startServices() / screen-manager.tsx setup flow:

• Check if HTTP server is already listening before starting it
• Check if MCP server process is alive before spawning a new one
• Connect to existing Nexus without re-running health init if nexusUrl resolves immediately

Impact

Reconnect time goes from ~8s → ~1-2s for the common case of re-attaching after detach or crash.