feat(mobile): render evidence levels and route actions in ask page

- normalize evidenceLevel/uncoveredPoints/nextActions from agent response
- GUIDANCE banner (reference experience, non-formal), PARTIAL uncovered
  points list, ROUTE compact action buttons (human help / direct channel /
  web research); legacy responses render unchanged
- one-tap human help on ROUTE: auto-submit down feedback (NO_ANSWER,
  fragment-less allowed by backend) then existing confirm + idempotent
  publish chain
- direct channel page supports ?channel= preselection

mobile-uni/src/types/api.ts

@@ -159,6 +159,9 @@ export interface KnowledgeSearchResponse {
   broadcastContext?: BroadcastKnowledgeContext | null;
   agentStatus?: AgentStatus;
   agentIntent?: AgentIntent;
+  evidenceLevel?: AgentEvidenceLevel | null;
+  uncoveredPoints?: string[];
+  nextActions?: NextAction[];
   sourceSummary?: AgentSourceSummary[];
   clarification?: AgentClarification | null;
 }
@@ -288,6 +291,16 @@ export interface AgentClarification {
   fields: string[];
 }
 
+export type AgentEvidenceLevel = 'FULL' | 'PARTIAL' | 'GUIDANCE' | 'ROUTE';
+
+export type NextActionType = 'HUMAN_HELP' | 'DIRECT_CHANNEL' | 'WEB_RESEARCH';
+
+export interface NextAction {
+  type: NextActionType | string;
+  channel?: string;
+  label?: string;
+}
+
 export interface AgentResponse {
   runId: string;
   conversationId?: string | null;
@@ -302,7 +315,9 @@ export interface AgentResponse {
   data?: unknown;
   actionDraft?: AgentActionDraft | null;
   clarification?: AgentClarification | null;
-  nextActions: string[];
+  evidenceLevel?: AgentEvidenceLevel | null;
+  uncoveredPoints?: string[];
+  nextActions?: NextAction[];
   keyPoints?: string[];
   cautions?: string[];
 }