Fix weighted evaluation score aggregation

[zjn20030811]

Summary

• include the skill-with-entity and exact-match metrics in total_score
• add regression tests for partial and perfect evaluation scores

Details

get_final_score computed four metrics and exposed all four in its return value,
but its weighted sum iterated only over skill_entity_scores. The two additional
metrics therefore had no effect and the reported maximum was capped at 80. This
change builds one score mapping before applying the existing weights, preserving
the public result fields and the configured 40/40/10/10 weighting.

Validation

• python -m pytest -q tests/test_evaluation_utils.py (2 passed)
• python -m compileall -q VLABench/evaluation/utils.py tests/test_evaluation_utils.py
• git diff --check

Fixes #77