Prefer case-local BFCL tool schemas

[SonAIengine]

Summary

• Prefer the current BFCL case row tool schema over the category-wide first definition when presenting retrieved tools to a model.
• Add a default schema-only argument-key instruction so models avoid undeclared argument names.
• Document the duplicate calculate_triangle_area schema collision and the Qwen 27B hardcase lift.

Validation

• poetry run ruff check .
• poetry run ruff format --check .
• poetry run pytest tests/test_bfcl_tool_selection_llm_loop.py tests/test_bfcl_tool_selection_sweep.py -q -> 25 passed
• poetry run pytest tests/ -q -> 961 passed, 3 skipped, 1 warning
• Qwen 27B cached smoke, simple_python_11: evaluator exact 1.0, predicted args base, height only
• Qwen 27B cached 6-case argument subset: evaluator exact 0.5, 3 pass, remaining tags optional_value_mismatch, percentage_scale_mismatch

Notes

• This is stacked on Tag BFCL argument mismatch causes #51 because GitHub still reports Tag BFCL argument mismatch causes #51 as open.

[SonAIengine]

Closing this internal stacked PR as superseded by the current main branch. The branch is preserved for history.