fix: respect zero policy sample thresholds

[Gsbreddy]

Summary

Fix policy sample-threshold overrides so explicit zero values are honored instead of falling back to workspace defaults.

Why

Policy permits min_baseline_runs, min_candidate_runs, and min_low_runs to be 0, but the diff logic used truthiness fallback. That made 0 behave like unset and could incorrectly block promotion on sample-confidence policy.

Changes

• Replace truthiness fallbacks with is not None checks in diff_releases.
• Add a regression test for zero policy sample thresholds.

Validation

• uv run python -c "from flightdeck.ledger import diff_releases; from flightdeck.models import Policy, PricingTable, WorkspaceConfig; policy=Policy(require_high_diff_confidence=True,min_candidate_runs=0,min_baseline_runs=0,min_low_runs=0); table=PricingTable(provider='openai', pricing_version='p', entries=[]); result=diff_releases(cfg=WorkspaceConfig(), policy=policy, baseline_events=[], candidate_events=[], baseline_pricing_table=table, candidate_pricing_table=table, window='7d'); print(result.confidence); print(result.policy.passed); print(result.policy.reasons)"
• uv run python -m pytest tests/test_ledger.py
• uv run python -m ruff check src tests
• uv run python -m pytest
• uv run python scripts/quickstart_smoke.py

Schema / Storage Impact

• None
• Schema change
• Storage change

Risk

Low. The change only distinguishes explicit zero thresholds from omitted thresholds.

Notes

Bug was reproduced before the fix: explicit zero thresholds returned LOW, False, and a default-threshold policy failure. After the fix, the same repro returns HIGH, True, and no policy reasons.