Graceful shutdown on Ctrl-C

[bug-ops]

Problem

Currently Ctrl-C breaks the agent loop immediately with no cleanup:

• No farewell message to the user
• MCP servers left orphaned (shutdown_all() exists but is never called)
• Mid-stream LLM responses are not cancelled — process hangs until the stream finishes

Scope

1. Add Agent::shutdown() method with MCP cleanup and farewell message
2. Call agent.shutdown() in main.rs after run() returns
3. Cancel-aware streaming loops (wrap stream.next() with tokio::select! on shutdown signal)

Files

• crates/zeph-core/src/agent/mod.rs
• crates/zeph-core/src/agent/streaming.rs
• src/main.rs

Reuse

• McpManager::shutdown_all() at crates/zeph-mcp/src/manager.rs:190
• shutdown_signal() helper at crates/zeph-core/src/agent/mod.rs:803