Subclass boto3.Session to use AWS_S3_ENDPOINT_URL

As of 0.1.13, the s3-tar library uses an environment variable
(`S3_ENDPOINT_URL`) to determine the AWS endpoint. See:
https://github.com/xtream1101/s3-tar/blob/0.1.13/s3_tar/utils.py#L25-L29.

To save BookWyrm admins from having to set it (e.g., through `.env`)
when they are already setting `AWS_S3_ENDPOINT_URL`, we create a Session
class that unconditionally uses that URL, and feed it to S3Tar.

bookwyrm/models/bookwyrm_export_job.py

@@ -3,6 +3,7 @@
 import logging
 from uuid import uuid4
 
+from boto3.session import Session as BotoSession
 from s3_tar import S3Tar
 from storages.backends.s3boto3 import S3Boto3Storage
 
@@ -25,6 +26,14 @@ from bookwyrm.utils.tar import BookwyrmTarFile
 logger = logging.getLogger(__name__)
 
 
+class BookwyrmAwsSession(BotoSession):
+    """a boto session that always uses settings.AWS_S3_ENDPOINT_URL"""
+
+    def client(service_name, **kwargs):
+        kwargs["endpoint_url"] = settings.AWS_S3_ENDPOINT_URL
+        return super().client(service_name, **kwargs)
+
+
 class BookwyrmExportJob(ParentJob):
     """entry for a specific request to export a bookwyrm user"""
 
@@ -211,6 +220,7 @@ class AddFileToTar(ChildJob):
                 s3_job = S3Tar(
                     settings.AWS_STORAGE_BUCKET_NAME,
                     f"exports/{filename}.tar.gz",
+                    session=BookwyrmAwsSession(),
                 )
 
                 # save json file