[DNM] feat(appkit): add AgentPlugin and ChatUIPlugin for LangChain/LangGraph agents (POC)

[smurching]

Summary

POC implementing the agent plugin plan. Promotes the LangChain/LangGraph agent pattern from app-templates/agent-langchain-ts into first-class AppKit SDK plugins.

• AgentPlugin (packages/appkit/src/plugins/agent/) — sets up a LangGraph ReAct agent (ChatDatabricks + optional MCP tools), streams responses in Responses API SSE format at POST /api/agent and registers POST /invocations at the app root (Databricks serving convention)
• ChatUIPlugin (packages/appkit/src/plugins/chat-ui/) — dynamically mounts a pre-built chat UI Express app as a sub-application
• Framework: added optional injectAppRoutes?(app) hook to Plugin base class; ServerPlugin calls it after mounting all /api/* routers
• New deps: @databricks/langchainjs, @langchain/core, @langchain/langgraph, @langchain/mcp-adapters

Usage

import { createApp, server, agent, chatUI } from '@databricks/appkit';

createApp({
  plugins: [
    server(),
    agent({
      model: 'databricks-claude-sonnet-4-5',
      systemPrompt: 'You are a helpful Databricks data assistant.',
      mcpServers: [/* optional MCP servers */],
    }),
    chatUI({ enablePersistence: false }),
  ],
});
// POST /api/agent — standard appkit endpoint
// POST /invocations — always registered (Databricks bundle compatibility)

Test plan

• TypeScript typecheck passes (pnpm typecheck)
• All 957 existing tests pass (pnpm test)
• Manual end-to-end: run dev playground with ENABLE_AGENT_EXAMPLE=true and test chat UI in browser

🤖 Generated with Claude Code