Prune subsumed BFCL partial tools

[SonAIengine]

Summary

• Keep raw retrieval evidence intact, but remove lower-level partial helpers from the model-facing BFCL candidate list when a richer case-local tool covers all query facets.
• Recover BFCL parallel_3 by presenting protein_info.get_sequence_and_3D without the sequence-only get_protein_sequence helper.
• Add focused regression coverage for the sequence + 3D model subsumption case.
• Update research docs with before/after Qwen artifacts and the combined route + subsumption smoke.

Validation

• poetry run ruff check .
• poetry run ruff format --check .
• poetry run pytest tests/test_bfcl_tool_selection_llm_loop.py -q -> 27 passed
• poetry run pytest tests/ -q -> 969 passed, 3 skipped, 1 warning
• Fresh Qwen before artifact: /tmp/gtc-bfcl-subsumption-before-parallel3-qwen.json -> call_count_mismatch
• Fresh Qwen after artifact: /tmp/gtc-bfcl-subsumption-after-parallel3-qwen.json -> evaluator exact 1.00
• Fresh Qwen two-case smoke: /tmp/gtc-bfcl-subsumption-route-parallel-qwen.json -> 2 / 2 pass, exact 1.00

Notes

• This does not alter the retrieved list; reports still show get_protein_sequence as retrieved for parallel_3, while tools_presented omits it.
• The next bottleneck is row-source repeated-call / argument mismatch, which is more model/schema-expression than retrieval loss.

[SonAIengine]

Closing this internal stacked PR as superseded by the current main branch.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