fix(agent): reject repeated completed tool call ids

[Sehlani042]

Summary

Fixes #872.

• track completed chat-tool and Responses-API tool call output IDs in the agent message history
• reject a model response that reuses an already-completed tool call ID before appending the assistant message or executing the shell action
• add a regression proving the repeated ID is surfaced as a format error and the duplicate shell action is not executed

Verification

• uv run pytest tests/agents/test_default.py::test_duplicate_completed_tool_call_id_is_not_executed_again tests/agents/test_default.py -q
• uv run pytest tests/models/test_actions_toolcall.py tests/models/test_test_models.py::test_response_api_model_format_observation -q
• uv run --extra dev ruff check src/minisweagent/agents/default.py tests/agents/test_default.py
• uv run --extra dev ruff format --check src/minisweagent/agents/default.py tests/agents/test_default.py
• git diff --check