Add unit tests for Formatter metrics

[Copilot]

Formatter.metrics() introduced in the parent PR had no test coverage for latency formatting, token extraction, or cost calculation, leaving those code paths unprotected against regressions.

Changes

• make_metrics_run helper — lightweight MagicMock factory that accepts extra, outputs, and a latency float to simulate start_time/end_time differences
• TestLatency — covers all three branches: sub-second (ms), seconds+ms, and minutes+seconds
• TestCosts — covers unknown model fallback, zero-token short-circuit, and non-zero token cost computation against a real model entry (openai/gpt-4o-mini)
• TestMetrics — covers token extraction, latency field formatting, cost lookup for a known model, and unknown-model cost fallback

All new tests follow existing conventions (no inline comments, class-per-feature, descriptive snake_case names).

class TestMetrics:
    def test_cost_for_known_model(self):
        run = make_metrics_run(
            extra={"ls_model_name": "openai/gpt-4o-mini"},
            outputs={"total_tokens": 2000, "cached_tokens": 1000},
        )
        result = Formatter.metrics(run)
        assert result["costs"] == {"cached": "0.000075", "total": "0.0012"}

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.