feat: implement core agent primitives (ReAct and Planner)

[Kaap10]

Description

This PR introduces the foundational agent architecture on top of the newly integrated multi-provider chat abstractions. It provides a robust, strictly-typed tool execution framework, a resilient ReAct loop capable of autonomous tool-use and error recovery, and a Planner for decomposing complex goals into sequential execution steps. This sets up the critical infrastructure needed for more advanced agentic workflows in the future.

Related issue

Fixes #288

Changes

• Implemented AgentTool class and a @tool decorator that leverages inspect.signature to automatically extract docstrings and generate strict JSON schemas (including robust support for __future__ stringified annotations).
• Implemented ReActAgent with synchronous (run) and asynchronous (arun) Think-Act-Observe loops, complete with step budget guards to prevent infinite loops and graceful observation-based error handling for failed tool executions.
• Implemented Planner agent to break down high-level objectives into strictly ordered PlanStep sequences using structured JSON outputs.
• Added a comprehensive suite of unit tests (tests/test_agents.py) verifying multi-step tool execution, async runtimes, budget exhaustion, unregistered tools, and error recovery using a deterministic offline fake model.

Testing

• Tests pass locally
• Ruff checks pass
• Documentation updated, if applicable (N/A for internal architecture, but docstrings are strictly enforced)

Checklist

• My changes are focused and relevant to this pull request.
• I have added or updated tests where appropriate.
• I have reviewed my changes for unrelated modifications.
• I have updated documentation where necessary.

[Kaap10]

Ready for review!