The email connector is providing the attachment data unextracted to the output connector as metadata attribute data. There are no transformation connectors that look at this metadata. Solr cell also probably does not handle binary in random metadata attributes the proper way.
The connector's attachment code therefore seems to be designed only to deal with textual attachments. The right solution is to have individual IDs for each attachment. But that would also require there to be a URL we could construct for each attachment. We could provide an additional URI template for attachments, but I'd wonder if your system has the ability to serve attachments by their own URLs. Please let me know if this would work and if so I can create a ticket and work on making these changes.