pr_review.py Passes a Copilot Round Whose Headline Names Findings Beside a Zero Open-Findings Count #2544

Problem

scripts/pr_review.py reads a Copilot ccr-overview-v2 review whose verdict headline names findings that no thread, no suppressed block and no open-findings count carries, and it reports the round as clean.

Constructed shape (the real one appeared on #2542):

<!-- ccr-overview-v2 -->

### 🔵 Needs a closer look

Resolve the two outstanding parser-retry and timeout-handling findings.

**0 open findings**

The digest for that round read threads=0 unresolved=0 suppressed=0 overview=0/0 shapes=ok, and wait exited 0. The headline is the one place the two findings exist, so a loop that trusts the digest merges over them. #2542 caught this only because the review body was read by hand, per the Merge Gate's "the matched review is read, not just counted".

This is a sibling of #2480, which covers the collapsible N open findings block. The class is the same, findings named in an overview section the reader does not parse, but the shape differs: here the count line says zero and the headline names findings.

Done when

• A round whose verdict headline is not the clean one, or whose headline sentence names outstanding findings, while open findings reads zero and no thread or suppressed block is on the head, is reported as its own reading rather than as a clean pass, and wait exits non-zero on it.
• A test feeds the constructed shape above.