fix(showcase): make safer promote artifact URI compatible with scenario replay

[w7-mgfcode]

Fresh-DB showcase_rich can hit a documented cascade where safer_promote_flow writes a placeholder artifact_uri that _parse_artifact_key cannot parse, causing downstream scenario_simulate_and_save to fail. This is documented in docs/_base/RUNBOOKS.md entry 18 and tolerated by the PRP-41 E2E test, but it should be fixed before the next release/demo.

Reproduction

1. Reset to a fresh DB: docker compose down -v && docker compose up -d && uv run alembic upgrade head.
2. Run /showcase with scenario=showcase_rich, Re-seed first + Reset database ticked.
3. Pipeline fails at step 16 (scenario_simulate_and_save) with Cannot parse artifact-key from artifact_uri: 'demo/safer-promote-placeholder.joblib'.

Scope options

• Option A: make safer_promote_flow (PRP-39 step in app/features/demo/pipeline.py) write a real-shape artifact_uri compatible with _parse_artifact_key (the V1 demo/{model_type}-model_{KEY}.joblib or V2 artifacts/models/model_{KEY}.joblib shapes).
• Option B: make scenario_simulate_and_save (or the underlying _parse_artifact_key in app/features/scenarios/) tolerate the placeholder and resolve the model artifact through another path (e.g. fall back to the alias's run row).

Acceptance

• Fresh-DB /showcase showcase_rich run reaches the PRP-41 agents + ops phases without the cascade firing.
• tests/test_e2e_demo.py::test_run_demo_showcase_rich_full_epic no longer needs the KNOWN_PREEXISTING_FAILURES = {\"scenario_simulate_and_save\"} tolerant-cascade handling — remove that set as part of this fix.
• No cross-slice imports into app/features/demo/ (the demo slice must keep going through HTTP / public surfaces, per repo conventions).

References

• docs/_base/RUNBOOKS.md § "Showcase page (`/showcase`) pipeline fails at step X" — entry 18 ("`scenario_simulate_and_save` step fails with `Cannot parse artifact-key from artifact_uri` (PRP-40, `showcase_rich` only)").
• PR feat(api,ui): showcase pipeline agent ops final polish (PRP-41) #323 — tests/test_e2e_demo.py::test_run_demo_showcase_rich_full_epic introduced the KNOWN_PREEXISTING_FAILURES allow-list that tolerates this cascade.
• Memory anchor: scenario-run-id-vs-registry-run-id — the /scenarios/* run_id is the forecast artifact key, NOT the registry's model_run.run_id; this ID-space confusion sits at the root of the cascade.