Durable workflow progression engine for Phase 25.

This is the single seam through which workflow runs move from active to waiting and from waiting/active to the next step. It evaluates the active step's normalized progress rules against the canonical prior delivery row inside one transaction, persisting the decision durably so that:

• wait_until rules write the run to :waiting with explicit status_reason: "waiting_for_step_progression" and a status_context map carrying rule_kind, anchor, anchor_delivery_id, anchor_delivery_status, anchor_timestamp, due_at, and to_step (D-01/D-13).
• on_outcome rules append a workflow_transition with reason progressed_on_delivery_outcome and the curated workflow outcome plus raw evidence facts (D-12), then advance the workflow run cursor to the target step and emit the next-step delivery through the canonical Chimeway.DeliveryPlanning.plan_next_step_delivery/3 seam (D-10).

Re-entry is duplicate-safe: if the run is no longer :active, if the prior delivery is not converged yet, if the curated mapper returns :not_branchable_yet, or if no progress rule matches, the engine returns {:noop, run, reason} without creating any new delivery rows or appending transitions. This is the ESC-03 contract.

All locking happens inside one Repo.transaction/1:

• the workflow run row is locked with FOR UPDATE first
• the active-step delivery row is then locked with FOR UPDATE

Threats covered:

• T-25-04 (tampering): outcomes are derived from canonical persisted rows only via ProgressionOutcome.from_delivery/2; the engine never branches from queue or in-flight job state.
• T-25-05 (repudiation): explicit status_reason and reason strings plus the curated status_context / transition context keys make the decision auditable from durable rows alone.
• T-25-06 (DoS / duplicate emission): noop short-circuits prevent retry storms from emitting duplicate next-step deliveries.