fix: thread AbortSignal through LLM calls so stop_goal cancels immediately

The agent loop checked signal.aborted only at the top of each iteration,
but the LLM fetch() call (which takes seconds) never received the signal.
Now the signal is passed to fetch() and checked after LLM errors and
before the inter-step sleep, so aborting takes effect mid-step.

server/src/agent/llm.ts

@@ -22,7 +22,8 @@ export interface LLMProvider {
   getAction(
     systemPrompt: string,
     userPrompt: string,
-    imageBase64?: string
+    imageBase64?: string,
+    signal?: AbortSignal
   ): Promise<string>;
 }
 
@@ -381,7 +382,8 @@ export function getLlmProvider(config: LLMConfig): LLMProvider {
     async getAction(
       systemPrompt: string,
       userPrompt: string,
-      imageBase64?: string
+      imageBase64?: string,
+      signal?: AbortSignal
     ): Promise<string> {
       const messages: Array<{ role: string; content: unknown }> = [
         { role: "system", content: systemPrompt },
@@ -418,6 +420,7 @@ export function getLlmProvider(config: LLMConfig): LLMProvider {
           max_tokens: 1024,
           response_format: { type: "json_object" },
         }),
+        signal,
       });
 
       if (!response.ok) {