review_log.t27: tests cannot tell str_starts_with from contains #7396

What

specs/queen/review_log.t27 classifies reviewer log lines. Two of its rules are prefix rules:

• a line that starts with reviewer: gh pr list is a listing error (LISTING_ERROR_504 when it also carries HTTP 504, otherwise LISTING_ERROR_OTHER);
• a line that starts with gh pr list -R: transient failure is a RETRY.

#7359 (b3617f2) replaced the body of the local str_starts_with helper, which had been a substring check, with a real prefix check. That was a correct fix, but none of the spec's 14 test blocks noticed: every fixture line that hits these rules starts with the needle, so the old substring body and the new prefix body give the same answers on all of them.

A test suite that passes for both the buggy and the fixed body does not pin the fix. If str_starts_with drifts back to str_contains, nothing goes red.

Fix

Add test blocks whose answer depends on the difference:

1. A line that contains reviewer: gh pr list but does not start with it (for example x reviewer: gh pr list -R -> HTTP 504: ...) classifies as OTHER, while the real prefixed line still classifies as LISTING_ERROR_504.
2. The same pair for the gh pr list -R: transient failure prefix: embedded is OTHER, prefixed is RETRY.
3. Direct assertions on str_starts_with (needle in the middle, needle at the start, needle longer than the haystack).

Acceptance

• t27c test-report --verbose specs/queen/review_log.t27 shows FAIL 0.
• Negative control: in a temporary copy whose str_starts_with body is replaced with return str_contains(haystack, needle);, t27c test-report shows FAIL >= 1.
• Seal refreshed with t27c seal --save.