Some of the places in which not_binary was being called were
not guarded in a way that made sure that the content-type being
sent to not_binary was just a media-type (without charset etc).
A parse_content_type method was extracted from the
extract_content_type method so that both a headers dict or a simple
header value could be parsed.
This seemed more generic and contractual than making not_binary
extract if required.
Fixes #158
4b82d0ddc7