Journal: durable run-journal + stall detector so the loop switches strategy instead of oscillating

[wesleysimplicio]

Parent epic: #64

Goal

Port simplicio-loop's run journal (scripts/loop_journal.py) to the piece pipeline: durable attempt memory per piece/campaign plus a stall detector, so when a piece keeps failing the same gate the loop changes strategy (new hook/angle, different provider via the routing matrix, different format, or escalate to human) instead of regenerating the same failing output forever.

Scope

• Append-only journal per piece (e.g. outputs/<client>/<date>/<piece-id>/journal.jsonl): attempt number, stage (copy/creative/critic/compliance/watcher-gate/tech-specs), gate result, failure signature, strategy used, provider resolved, cost estimate.
• Stall detection: N consecutive failures on the same gate with the same failure signature → the loop must not retry the same strategy; it must pick an untried one from a strategy ladder or route the piece to review with the journal attached.
• Strategy ladder is declarative (e.g. rewrite-hook → change-format → change-provider → human-review) and provider-agnostic — switches go through lib/router.ts, never name a vendor.
• Journal feeds data/runs.jsonl aggregation and the learnings lane (Analytics: score organic performance across social, Reddit, forums, and dev portals #54 replay, lib/promotion/learnings.ts).

Acceptance criteria

• E2E test: a mocked piece failing compliance 3× with the same violation triggers a strategy switch (visible in the journal) instead of a 4th identical retry.
• E2E test: exhausted ladder routes the piece to human review with blocked status — never silently drops it or fake-passes.
• Journal survives process restart (durable on disk, resume picks up attempt count).
• selftest covers signature hashing and stall threshold logic.