fix: fix PyTorchModel deployment crash on Windows (#1212)

Starting from version `1.2` deployment of PyTorchModel always
initiates repacking when uploading the code. It is done by
calling `_upload_code` with `repack=True` parameter.
On Windows, this will lead to a crash, because
`os.path.join` method used to generate `repacked_model_uri` will
use backslashes instead of slashes. So `utils.repack_model` will fail to parse it.

Author: Serhiy-Shekhovtsov

src/sagemaker/model.py

@@ -817,7 +817,7 @@ def _upload_code(self, key_prefix, repack=False):
 
         if repack:
             bucket = self.bucket or self.sagemaker_session.default_bucket()
-            repacked_model_data = "s3://" + os.path.join(bucket, key_prefix, "model.tar.gz")
+            repacked_model_data = "s3://" + "/".join([bucket, key_prefix, "model.tar.gz"])
 
             utils.repack_model(
                 inference_script=self.entry_point,