Packet residuals contradict insufficient_evidence low-confidence count

[pengfei-threemoonslab]

Found during P0.3 public-demo advisory release-evidence rehearsal (#100).

The scan of openai/openai-agents-python/examples/customer_service produced:

• release_decision.decision = insufficient_evidence
• reason: evidence coverage below threshold due to 2 low-confidence tools
• release_decision.evidence_coverage.low_confidence_tool_count = 2

But packet.md §10 rendered: Low-confidence tool extractions: none.

Why it matters: the packet is the reviewer-facing artifact. If §1 and §10 disagree, reviewers cannot tell whether the evidence gap is real.

Acceptance criteria:

• Packet residuals render the same low-confidence tool count used by release_decision.evidence_coverage.
• Add/adjust a regression test that covers insufficient_evidence from low-confidence SDK extractions and packet §10 output.