refactor(models): Refine MessageAgentThought SQLAlchemy typing

[tomerqodo]

Benchmark PR from agentic-review-benchmarks#11

[augmentcode]

🤖 Augment PR Summary

Summary: Refactors the MessageAgentThought model to use richer SQLAlchemy typing and more explicit defaults.

Changes:

• Updates MessageAgentThought to inherit from TypeBase and adds/aligns typed Mapped[...] columns (including Decimal pricing fields).
• Adjusts agent thought creation to use Decimal values and CreatorUserRole for created_by_role.
• Makes thought appending in save_agent_thought robust to NULL existing values.
• Refines tool input/observation parsing in agent history reconstruction to better handle missing/malformed JSON.
• Updates integration tests to stop passing explicit id values for MessageAgentThought.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

tool_responses is parsed above but this ToolPromptMessage uses tool_inputs, which looks like it will surface the tool inputs as the tool output in history. This also makes tool_responses unused, so consider switching this to tool_responses.get(...) (stringified) to preserve the actual observation/result.

_{🤖 Was this useful? React with 👍 or 👎}

[augmentcode]

In create_agent_thought, answer_unit_price=Decimal("0.001") with answer_price_unit=Decimal(0) seems inverted compared to the message fields and the DB defaults (price unit defaulting to 0.001). If this record is read before llm_usage is applied, these initial values could misrepresent pricing.

_{🤖 Was this useful? React with 👍 or 👎}