Uptake bc-eval 0.3.12 and drop the evaluator CAPI taxonomy shim

[Copilot]

bc-eval 0.3.12 sends the M365 LLM API taxonomy/CoS headers itself, so the workaround added in #756 for the evaluator/judge path is no longer needed.

Version bumps

• .github/workflows/summarize-results.yml: bc-eval[capi]==0.3.8 → 0.3.12
• .github/workflows/bcal-evaluation.yml: bc-eval[capi]==0.3.7 → 0.3.12

Cleanup

• evaluator/scores.py: removed the sys.path hack and _install_capi_taxonomy() call; the file is now just the evaluator classes.
• Deleted evaluator/_capi_taxonomy.py and tests/test_capi_taxonomy_evaluator.py.

Unchanged

• The taxonomy shim in src/bcbench/agent/bcal/bc_eval_capi_bridge.py and the CAPI_TAXONOMY_* env vars in bcal-evaluation.yml are kept — the bridge patches CapiModel for the bcal target-model path and still needs them.

[Copilot]

Pull request overview

Updates bc-eval to use its native CAPI taxonomy headers and removes the obsolete evaluator shim.

Changes:

• Bumps bc-eval to 0.3.12 in evaluation workflows.
• Removes the evaluator-side taxonomy patch and tests.
• Retains the separate bcal bridge taxonomy handling.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
.github/workflows/summarize-results.yml	Updates bc-eval to 0.3.12.
.github/workflows/bcal-evaluation.yml	Updates the bcal bridge environment dependency.
evaluator/scores.py	Removes shim initialization.
evaluator/_capi_taxonomy.py	Deletes the obsolete taxonomy shim.
tests/test_capi_taxonomy_evaluator.py	Deletes tests for the removed shim.