[Feature]: Add Output Schema

[0xKoller]

Problem statement

Right now, when defining a tool in xmcp, the output schema is not supported.
Adding support for an output schema would significantly improve compatibility with LLM tool-calling (OpenAI, Anthropic, etc.), which benefits greatly from explicit schemas.

Proposed solution

• Automatically infer the output schema from the handler function's return type (using TypeScript reflection, Zod inference helpers, or a similar mechanism).
• Assign it to the tool's outputSchema property if no explicit schema was provided by the developer.

Package

xmcp (core framework)