[Example] Add MCP server example for AI agent integration

### Motivation

The `examples/` directory shows basic producer/consumer, async, and RPC patterns, but there is no example demonstrating integration with AI agents.

The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) is the emerging standard for connecting AI tools (Cursor, Claude Desktop, Windsurf, etc.) to external data sources. An MCP server example would allow developers to interact with Pulsar topics through natural language, bridging messaging infrastructure and the AI agent ecosystem.

### Proposed Change

Add `examples/mcp_server.py` that exposes three tools via the official MCP Python SDK (`pip install mcp`):

| Tool | Description |
|------|-------------|
| `pulsar_publish` | Publish a message to a topic (supports properties, partition key, delayed delivery) |
| `pulsar_consume` | Consume and auto-acknowledge messages from a subscription |
| `pulsar_peek` | Read messages using a Reader without affecting consumer state |

The example is a standalone script—`mcp` is only required at runtime for this example and is **not** added to the project's `setup.py` dependencies.

### Configuration

Once configured in an MCP-compatible client (e.g. Cursor `~/.cursor/mcp.json`), users can produce and consume Pulsar messages through natural language.

```json
{
"mcpServers": {
"pulsar": {
"command": "python3",
"args": ["examples/mcp_server.py"],
"env": {
"PULSAR_BROKER_URL": "pulsar://localhost:6650"
}
}
}
}
```

I'm happy to submit a PR for this.