fix: throw error not lorem when moderation fails

Instead of switching to test mode, which responds with "Lorem Ipsum"
sample output to test the API, throw an appropriate error message for
content moderation failures. This is especially useful when OpenAI's
moderation is way way way way too aggressive as this confuses users.

src/backend/src/api/APIError.js

@@ -509,6 +509,12 @@ module.exports = class APIError {
             status: 400,
             message: ({ engine, valid_engines }) => `Invalid engine: ${quot(engine)}. Valid engines are: ${valid_engines.map(quot).join(', ')}.`,
         },
+        
+        // Abuse prevention
+        'moderation_failed': {
+            status: 422,
+            message: `Content moderation failed`,
+        },
     };
 
     /**

src/backend/src/modules/puterai/AIChatService.js

@@ -324,6 +324,7 @@ class AIChatService extends BaseService {
 
                 if ( ! test_mode && ! await this.moderate(parameters) ) {
                     test_mode = true;
+                    throw APIError.create('moderation_failed');
                 }
 
                 if ( ! test_mode ) {
@@ -622,6 +623,7 @@ class AIChatService extends BaseService {
     * Returns true if OpenAI service is unavailable or all messages pass moderation.
     */
     async moderate({ messages }) {
+        if ( process.env.TEST_MODERATION_FAILURE ) return false;
         for ( const msg of messages ) {
             const texts = [];