refactor(models): Refine MessageAgentThought SQLAlchemy typing

[tomerqodo]

Benchmark PR from qodo-benchmark#425

[augmentcode]

🤖 Augment PR Summary

Summary: This PR refactors the MessageAgentThought SQLAlchemy model to use stronger typing and dataclass-friendly defaults.

Changes:

• Converted MessageAgentThought to a typed TypeBase mapping with explicit Mapped[...] annotations and defaults (including Decimal pricing fields).
• Updated agent runner persistence to align with the refined model types (Decimal values, explicit tool_process_data, and enum-based created_by_role).
• Hardened agent-thought text concatenation to handle nullable thought fields.
• Adjusted tool input/observation parsing when reconstructing agent history (handling empty payloads and malformed JSON).
• Updated integration tests to stop passing explicit id values now that IDs are generated by the model mapping.

Technical Notes: The model now defaults nullable fields to None and uses Decimal-typed columns for price/unit fields, which may affect any code that relied on implicit integer defaults.

_{🤖 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]

In create_agent_thought, answer_unit_price/answer_price_unit look swapped vs the model defaults (*_price_unit is non-null and defaults to 0.001). Setting answer_price_unit=Decimal(0) could lead to incorrect cost calculations (or division-by-zero assumptions elsewhere) if this row is read before llm_usage is applied.

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

[augmentcode]

ToolPromptMessage.content is built from tool_inputs here, so the reconstructed tool message history will contain inputs rather than tool observations; also tool_responses becomes effectively unused. This likely breaks downstream behavior that expects the tool-response payload in the tool message.

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