[Bug]: Replace fake llm.summarize tool with real example in composite tools docs

[danbarr]

Issue

In the composite tools documentation (docs/toolhive/guides-vmcp/composite-tools.mdx:82), we use a hypothetical llm.summarize tool in the example:

- id: summarize
  tool: llm.summarize # Hypothetical backend - replace with your actual LLM server
  arguments:
    text: '{{.steps.fetch_content.output.content}}'
  dependsOn: [fetch_content]

This is confusing for users because:

1. It's not a real tool they can use
2. It gives the impression this is something that exists
3. Users trying to follow the example will get errors

Proposed solution

Replace the example with a real MCP server tool that users can actually deploy and test with. Options:

• Use a different real example that doesn't require an LLM
• Document how to set up a real LLM MCP server and use that
• Use a simpler two-step example with tools that definitely exist (e.g., fetch → create GitHub issue)

Context

From PR #326 review comment by @danbarr

[jhrozek]

I think the perplexity MCP server could do the trick here, but we might need to fix stacklok/toolhive#2921 first - I actually wanted to use the perplexity MCP and ran into the issue on Friday.