reduce memory usage when # of emails > chunk size

[pawl-bb]

Currently CeleryEmailBackend.send_messages works like this:

1. All e-mails are serialized with email_to_dict and stored in a list
2. The list of serialized e-mails is split into chunks
3. A celery task is created for each chunk

Serializing all the e-mails at once can use lots of memory if you have lots of e-mails with attachments.

To improve memory efficiency, this PR changes CeleryEmailBackend.send_messages to run email_to_dict on one chunk of serialized e-mails at a time.

Benchmarks (on 80 emails with a 5 mb attachment):

• Before: 1028 mb (peak memory usage)
• After: 269 mb