feat(grug): every review says which commits it read

[quadseven]

Why

"Looked at: <range>" was emitted only when a Living Hunt delta existed. A full base..head review - the common path, and every first review of any PR - answered "what did you actually read" with nothing.

That is the first question a skeptical reader asks about a review, and on most PRs it had no answer.

Summary

Both cases disclose now, and both name a range rather than a single commit: base..head is what the author pastes into git diff to see exactly what Grug saw.

One implementation, board.review_scope_line. The full-diff branch was briefly written twice - once there, once inline in dispatch.py with its own wording and its own head_sha[:7] - so the same fact appeared in two vocabularies on the same surface.

Acceptance criteria

• A full base..head run discloses its scope - test_stack_discloses_scope_on_a_full_diff_run_too
• A Living Hunt delta run still discloses its range unchanged
• The disclosure is a RANGE, not one commit - asserted as base987..c0ffee1
• No dangling "Looked at:" when there is no sha to name - test_stack_detail_without_a_head_sha_still_renders
• One implementation of the fact, not two
• Suites: 1883 webhook, 288 api, 24 infra

Size

Size: S

Out of scope

#673's in-progress note is deliberately NOT here. It was implemented, reviewed and removed before this PR rather than shipped. Two-axis review found it net-negative:

1. It omitted app_id, so find_board could match any comment quoting the board marker - including a human's. And because the key is elder, board_client's legacy branch (if key != "elder") discards existing text rather than preserving it, so a human quoting the marker would have had their comment replaced wholesale.
2. Nothing cleared it. The only terminal header write is gated on not check_publish_failed, so a publish failure, a consumer crash into the DLQ, post-settle ineligibility, or the fail-open path would each leave "Wait" on the board permanently - with the prior findings already overwritten. Strictly worse than the staleness it was meant to fix.

Re-filed with both as hard acceptance criteria.

#673 item 3 (disclose tier/arms/model) is also out. Per operator decision it moves to the dashboard rather than the author's PR comment - it is useful to whoever runs Grug, not to whoever wrote the PR.

Refs #673