fix: retry all SQS messages on unhandled errors instead of silently dropping them

When the scale-up lambda encounters an unhandled error (e.g. SSM
ThrottlingException during registration token creation), the catch block
returned an empty batchItemFailures array. With ReportBatchItemFailures
enabled, this tells SQS that all messages were processed successfully,
permanently deleting them from the queue.

This causes queued GitHub Actions jobs to be silently lost — they never
get a runner and remain stuck in 'queued' state indefinitely.

The fix returns all message IDs as batch item failures on unhandled
errors, so SQS retries them after the visibility timeout.

Author: Snojo

lambdas/functions/control-plane/src/lambda.ts

@@ -55,9 +55,11 @@ export async function scaleUpHandler(event: SQSEvent, context: Context): Promise
       batchItemFailures.push(...e.toBatchItemFailures(sqsMessages));
       logger.warn(`${e.detailedMessage} A retry will be attempted via SQS.`, { error: e });
     } else {
-      logger.error(`Error processing batch (size: ${sqsMessages.length}): ${(e as Error).message}, ignoring batch`, {
-        error: e,
-      });
+      batchItemFailures.push(...sqsMessages.map(({ messageId }) => ({ itemIdentifier: messageId })));
+      logger.error(
+        `Error processing batch (size: ${sqsMessages.length}): ${(e as Error).message}, all messages will be retried via SQS.`,
+        { error: e },
+      );
     }
 
     return { batchItemFailures };