Summary

Populates metric.score_breakdown in StepEfficiencyMetric after each evaluation. The breakdown maps each agent step to its necessity verdict and reason, giving users per-step visibility into which actions were efficient and which were redundant.

Changes

• deepeval/metrics/step_efficiency/schema.py — added StepAnalysis model and an optional steps: List[StepAnalysis] field to EfficiencyVerdict
• deepeval/metrics/step_efficiency/template.py — extended the prompt to request per-step analysis alongside the overall score
• deepeval/metrics/step_efficiency/step_efficiency.py — populate self.score_breakdown from efficiency_verdict.steps in both sync and async paths
• tests/test_metrics/test_step_efficiency_metric.py — added TestStepEfficiencyScoreBreakdown with two no-API-key unit tests verifying correct score_breakdown population

Test plan

• python -m pytest tests/test_metrics/test_step_efficiency_metric.py::TestStepEfficiencyScoreBreakdown -v — passes (no API key required)
• python -m black --check deepeval/metrics/step_efficiency/ — passes
• python -m ruff check deepeval/metrics/step_efficiency/ — passes

Example

metric = StepEfficiencyMetric()
metric.measure(test_case)

print(metric.score)          # e.g. 0.5
print(metric.score_breakdown)
# {
#   "fetch_data":     {"necessary": True,  "reason": "Required to retrieve input."},
#   "extra_llm_call": {"necessary": False, "reason": "Duplicate; first response was sufficient."},
# }

Fixes / Closes

Closes #2313