fix aiohttp crashing on gzipped HEAD responses

 * This is a hack.  aiohttp tries to unzip the content body for HEAD
   responses when the Content-Encoding header is present.  Since HEAD
   responses don't have bodies, this crashes with a content encoding
   error.  This bug has been fix in aiohttp v0.22, but we are not yet
   ready to upgrade that.

mfr/providers/osf/provider.py

@@ -7,6 +7,7 @@
 
 import furl
 import aiohttp
+from aiohttp.errors import ContentEncodingError
 
 from waterbutler.core import streams
 
@@ -69,11 +70,14 @@ async def metadata(self):
             # To make changes to current code as minimal as possible
             try:
                 metadata = {'data': json.loads(metadata_request.headers['x-waterbutler-metadata'])['attributes']}
+                await metadata_request.release()
             except KeyError:
                 raise exceptions.MetadataError(
                     'Failed to fetch metadata. Received response code {}'.format(str(metadata_request.status)),
                     code=400)
-        await metadata_request.release()
+            except ContentEncodingError:
+                pass  # hack: aiohttp tries to unzip empty body when Content-Encoding is set
+
         self.metrics.add('metadata.raw', metadata)
 
         # e.g.,