fix(ai): completely remove thinkingConfig from Gemini payload to prevent 400 errors

lib/ai/gemini-chat.ts

@@ -113,9 +113,7 @@ export async function callGeminiChat(opts: {
         config.systemInstruction = opts.systemPrompt;
     }
 
-    if (opts.includeThoughts) {
-        config.thinkingConfig = { thinkingBudget: 1024 };
-    }
+    
 
     const fns = toGeminiFunctions(opts.tools ?? []);
     if (fns) config.tools = fns;
@@ -174,7 +172,7 @@ export async function* streamGeminiChat(opts: {
     const config: any = {
       temperature: opts.temperature ?? 0.7,
       maxOutputTokens: 8192,
-      thinkingConfig: { thinkingBudget: 1024 },
+      
     };
 
     if (opts.systemPrompt) {

lib/ai/gemini-client.ts

@@ -86,11 +86,7 @@ export class GeminiLlmClient implements LlmClient {
       config.systemInstruction = `${args.systemPrompt}\n\nIMPERATIVE: Respond ONLY with this exact JSON format:\n${JSON.stringify(exampleJson)}\n\nDo NOT add any other fields.`;
     }
 
-    if (args.thinking_config) {
-      config.thinkingConfig = {
-        thinkingBudget: 1024,
-      };
-    }
+    
 
     const run = async (currentContents: any[]) => {
       console.log(`[GoogleGenAI] generateContent with ${this.model}`);