Add: minimal tool-use agent loop notebook

[mnifzied-create]

What this adds

A from-scratch teaching notebook — tool_use/minimal_agent_loop.ipynb — that implements the core tool-use agent loop by hand with the Anthropic SDK: call the model, run the tool(s) it requests (including several in one turn), feed the results back as a single tool_result user turn, and repeat until stop_reason is no longer tool_use.

Why

The existing tool_use/ notebooks show specific tools and full agents, and claude_agent_sdk/ covers higher-level agents — but there isn't a minimal, framework-free walkthrough of the loop itself, the concept every agent is built on. This fills that gap in ~30 readable lines, with markdown on the four details people most often get wrong: appending the assistant turn verbatim, one tool_result per request matched by tool_use_id, all results in a single user message, and a max_turns safety belt.

Quality

• Current model id (claude-sonnet-4-6); code-cell outputs left empty for CI execution.
• Loop logic mock-verified end-to-end (multiple tool_use blocks in one turn, correct id matching, clean termination on end_turn).
• ruff check / ruff format --check clean; nbformat valid; registry.yaml + authors.yaml validate against the repo schemas.

Happy to adjust placement, scope, or framing — thank you for considering it!