feat(opal): tint a guess red once its guesser ran out of guesses

[vEnhance]

Enhancement, following up on #563.

What this changes

Yellow says a guess led nowhere, but it doesn't separate someone who wandered off from someone the puzzle actually shut out — and the second is the sharper signal, since they wanted the answer and the guess limit took it away.

This adds a red tint (table-danger) for a row whose guesser has no correct answer for that puzzle and has spent its whole guess limit. It slots in between the finisher and unsolved branches:

testsolver → blue
finished the hunt → green
unsolved AND out of guesses → red   ← new
unsolved → yellow
otherwise → plain

So the hunt-wide tints still win, and every red row is one that would otherwise have been yellow. Given how few yellow rows there are in practice, this should be a rare and meaningful marker rather than a wall of color.

Keeping "out of guesses" honest

_eligibility was the only definition of which guesses count against the limit — not correct, not close, not excused. The log has to agree with it, or a row goes red while the puzzle page is still happily taking answers. Both now read one COUNTS_AGAINST_GUESS_LIMIT Q object.

test_guess_log_out_of_guesses_matches_puzzle_page pins the two together directly: it walks a guesser to the limit (with a close guess and an excused guess mixed in, since neither should count) and asserts can_attempt flips from True to False in the same step the tint goes from yellow to red.

Query cost

Still one query for the standings, not two. Rather than add a second query to count spent guesses, _standings now groups every guess on the hunt by (guesser, puzzle) and reads two filtered aggregates off each group:

.annotate(
    first_correct=Min("created_at", filter=Q(is_correct=True)),
    num_counted=Count("pk", filter=COUNTS_AGAINST_GUESS_LIMIT),
)

That carries all four facts a standing holds — solved set, exhausted set, testsolver, finisher — with identical semantics to the previous correct-attempts-only query. Measured at 2, 50, and 100 rows, all four logs are unchanged: 10 (hunt log), 9 (recent activity), 12 (per-puzzle), 12 (per-user).

Testing

test_guess_log_row_styling now also walks Bob to the meta's guess limit and checks his rows on it turn red while his rows on the puzzle he did solve stay plain, then that finishing the hunt flips everything green.

Full suite passes (424), make check clean.

---

Generated by Claude Code