Add back logic to store parsed Gemini output in model response

This was accidentally removed in commit aa48ca4 and was preventing agent
responses from being properly parsed (in particular, agent participant
responses from being correctly parsed into stage answers).

Author: vivtsai

functions/src/api/gemini.api.ts

@@ -15,6 +15,7 @@ import {
   StructuredOutputSchema,
   ModelResponseStatus,
   ModelResponse,
+  addParsedModelResponse,
 } from '@deliberation-lab/utils';
 
 const GEMINI_DEFAULT_MODEL = 'gemini-2.5-flash';
@@ -164,6 +165,7 @@ export async function callGemini(
   prompt: string | Array<{role: string; content: string; name?: string}>,
   generationConfig: GenerationConfig,
   modelName = GEMINI_DEFAULT_MODEL,
+  parseResponse = false, // parse if structured output
   safetySettings?: SafetySetting[],
 ): Promise<ModelResponse> {
   const genAI = new GoogleGenAI({apiKey});
@@ -253,6 +255,10 @@ export async function callGemini(
     imageDataList: imageDataList.length > 0 ? imageDataList : undefined,
   };
 
+  if (parseResponse) {
+    return addParsedModelResponse(modelResponse);
+  }
+
   return modelResponse;
 }
 
@@ -318,6 +324,7 @@ export async function getGeminiAPIResponse(
       promptText,
       geminiConfig,
       modelName,
+      structuredOutputConfig?.enabled,
       safetySettings,
     );
     // eslint-disable-next-line @typescript-eslint/no-explicit-any