synthetic-toy → 200 cases: outcome leaderboard row lands; all 5 boards real

[protosphinx]

Stacked on top of #11 (conformance). Merge order: #2 → #3 → #4 → #5 → #6 → #7 → #8 → #9 → #10 → #11 → this.

Summary

• Bumps synthetic_log() default from 50 → 200 cases. The chronological tail of the bigger log naturally contains positive delivery_confirmed cases, so the outcome task finally has a real leaderboard entry.
• All 4 existing reference predictions regenerated; AUC entry added for outcome. Five boards now verify under --all --verify.

What's new

• pm_bench/_synth.synthetic_log(n_cases=200, seed=42) — same path distribution, more cases. Test partition: 158 prefixes, 45 positives.
• 5th leaderboard board: leaderboard/outcome/synthetic-toy.json with prior-ref entry — AUC 0.6319, n_pos 45/158. Real floor for any temporal model on the outcome task.
• _rescore_outcome + _outcome_truth_for_dataset wired into pm_bench/leaderboard.py.
• All 4 existing entries regenerated:
  • markov-ref: top-1 0.9304 (was 0.9756 on 50 cases — Markov is harder when the test set has more variety)
  • mean-ref: MAE 1.3481 days
  • mean-wait-ref: NDCG@10 0.9911 over 9 transitions
  • dfg-ref: F = 1.0 (200 cases → both partitions observe every path-graph edge — the cleaner conformance result)
• registry.yml synthetic-toy row updated (cases 200, events 965).
• STANDINGS.md regenerated; README.md, STATUS.md updated.
• Tests adjusted to the new numbers; the n_pos=0 by accident comments are gone.

Smoke

$ pm-bench leaderboard --all --verify
bottleneck/synthetic-toy: OK - 1 entry(ies)
conformance/synthetic-toy: OK - 1 entry(ies)
next-event/synthetic-toy: OK - 1 entry(ies)
outcome/synthetic-toy: OK - 1 entry(ies)
remaining-time/synthetic-toy: OK - 1 entry(ies)

Test plan

• pytest -q — 109 passed (was 108 on PR v0.3 closed: conformance task — DFG fitness × precision → F #11; +1 for the new outcome-board verify test)
• ruff check pm_bench tests — clean
• pm-bench leaderboard --all --verify exits 0 across all 5 boards
• STANDINGS.md staleness canary green

Roadmap impact

• Closes the long-standing "outcome row waits on a dataset whose test split has both classes" caveat. Synthetic-toy is now self-sufficient for all 5 v0 tasks.
• Numbers will move when a real BPI dataset gets pinned, which is fine — that's the whole point of having drift detection in CI.

[protosphinx]

Merged into main as part of the audit-cleanup stack (commit 9c00b47). The full content of this PR is now on main.