Updates and wiring

feat(ai): integrate prediction and optional ML scoring; fix AlertsPanel crash

- dispatcher:
  - Add DISPATCHER_ENABLE_ML_SCORING flag (default: false).
  - Implement fetchPredictionConfidence() calling /api/v1/prediction/simulate and
    combine ML confidence with incoming alert confidence (max, clamped [0,1]).
  - Use effectiveConfidence in policy evaluation to influence mission creation.

- console:
  - summitClient.getPrediction(): point to /api/v1/prediction/simulate with simplified input.
  - AlertsPanel: harden against missing fields (type/severity/location/timestamp) to
    prevent runtime crash (toUpperCase on undefined).
  - Wire SIMULATE button to call summitClient.getPrediction() (logs result for now).

- notes:
  - ML scoring is feature-flagged; no behavior change unless enabled.
  - Follow-ups: render prediction isochrones on the map; surface ML rationale in UI.

- files:
  - apps/mission-dispatcher/src/index.ts
  - apps/console/src/lib/summitClient.ts
  - apps/console/src/components/panels/AlertsPanel.tsx

- validate:
  - Start API GW exposing /api/v1/prediction/simulate.
  - Run dispatcher with DISPATCHER_ENABLE_ML_SCORING=true and DISPATCHER_REQUIRE_CONFIRM=true.
  - Publish a mock alert; confirm mission creation and no UI crashes.

Author: kpres12