feat(core): introduce ToolCall.arguments_dict and ToolResult.from_value

[seanbrar]

Summary

This PR adds ergonomics and safety improvements for handling tool arguments and results, and guarantees structured output serialization:

• Adds ToolCall.arguments_dict() to strictly validate and retrieve parsed JSON-object arguments, failing fast with a ConfigurationError on malformed JSON or non-object arguments.
• Adds ToolResult.from_value() constructor to automatically serialize non-string values to compact JSON and handle error payloads.
• Updates Output.to_jsonable() to properly serialize Pydantic BaseModel structured outputs using value.model_dump(mode="json").
• Modernizes docs/agent-loop.md to teach these recommended v2 patterns.

Related issue

None

Test plan

Run the local checking tool to execute all tests:

just check

Verified that 423 tests passed, including new unit tests covering:

• JSON parsing and object-shape validation for arguments_dict().
• Automatic serialization of generic JSON value types and strings for from_value().
• Clean serialization of Pydantic models in to_jsonable().

Notes

None

---

• PR title follows conventional commits
• just check passes
• Tests cover the meaningful cases, not just the happy path
• Docs updated (if this changes public API or user-facing behavior)

[codecov]

Codecov Report

❌ Patch coverage is 86.95652% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/pollux/interaction/tools.py	88.23%	2 Missing ⚠️
src/pollux/interaction/output.py	83.33%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!