A linked repo's truncated patch can still reach a verdict unread #602

Problem

#437 closed the case where a cut changed-file patch in the PRIMARY diff could reach a PASS unseen: budget_diff names the cut paths, InspectionTracker requires each to be referenced, and reviewer.py:3317-3319 turns a miss into a no-verdict instead of a decision.

A linked repo's cut paths are not covered by any of that.

review/reviewer.py:635, in _linked_repos_review_section:

diff, total, _cut_paths = _git_diff(lpath, lbefore, "HEAD")

The third value is bound to _cut_paths and never used. The only producer of required_inspections is reviewer.py:2797, which passes the PRIMARY diff's cut_paths alone. So a linked repo whose patch is truncated has no path recorded as required, and the reviewer can reach a verdict without the cut file ever coming up — exactly the failure #437 describes, in a narrower place.

This is recorded as a known scope limit in InspectionTracker's own docstring (review/diff_coverage.py), which says widening the check "belongs with whoever gives linked repos coverage that matters". Filing it so that is tracked rather than living only in a docstring.

Acceptance criteria

• A review whose LINKED repo diff is truncated records that repo's cut paths as required inspections, and a verdict reached without referencing them is rejected the same way the primary diff's are. A test fails on main.
• The rejection message distinguishes a linked repo's path from the primary repo's, so the reason names which repo was not read.
• A review with no linked repos, and one whose linked-repo diff is not truncated, behave exactly as they do today.
• The primary diff's existing coverage behaviour is unchanged, pinned by the tests already covering it.

Reported by @EllaSharakanski as part of #437; split out because #437's own criteria are met on main.