feat: add `ToolDefinition#output_schema`

- [x] I have looked for existing issues (including closed) about this

## Feature Request

In `ToolDefinition` it would be nice to define a output JSON schema as well, such that LLM can understand the values returned from the tool

### Motivation

Have the LLM better understand the output returned from a tool

### Proposal

I think this should work for OpenAI, Gemini at least

Look for the ToolDefinition struct in the codebase, likely in a core module. Understand how input_schema is defined and used. The change involves adding an output_schema field and ensuring it's integrated with LLM providers like OpenAI and Gemini. Check existing tool usage and tests to see how output is currently handled.

Scritto dal modello di indicizzazione a partire dal testo della issue.