Export BFCL failure tag subsets

[SonAIengine]

Summary

• preserve non-empty BFCL model-loop failure_tags in extracted failure cases
• emit tag_.txt subset files from hard-case bundles
• document tag-based hard-case loops for near_duplicate_tool_surface experiments

Validation

• poetry run pytest tests/test_bfcl_tool_selection_failures.py tests/test_bfcl_tool_selection_hard_cases.py -q
• poetry run pytest tests/test_bfcl_tool_selection_hard_cases.py tests/test_bfcl_tool_selection_sweep.py tests/test_bfcl_tool_selection_llm_loop.py -q
• poetry run ruff check .
• poetry run ruff format --check .
• poetry run pytest tests/ -q
• REPORT=/tmp/gtc-bfcl-qwen027-cohesive-guard-limit25-equivalence.json OUT_DIR=/tmp/gtc-bfcl-qwen027-tag-subsets TOOL_SOURCES=retrieved REPORT_TOP_KS=5 TOP_K=5 FAILURE_CATEGORIES=retrieval_miss,candidate_ambiguity,argument_value_mismatch,call_count_mismatch make bfcl-hard-cases

Cached artifact produced tag_near_duplicate_tool_surface.txt with:

• simple_python_6
• simple_python_12
• simple_python_21
• simple_python_22

[SonAIengine]

Closing this internal stacked PR as superseded by the current main branch.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