There are several places that call _EncodeLOB; most are careful to pass
bytes, not characters:

  1. RT::Attachment->Create takes a MIME::Entity; while the transfer
     encoding will have been decoded, ->bodyhandle->as_string does not
     decode bytes into characters.

  2. ObjectCustomFieldValues from file uploads; these are always left as
     bytes.

  3. ObjectCustomFieldValues from Content which is too long; Content is
     passed as characters.

The one codepath which currently might pass characters, and not bytes,
is the third possibility.  While the Mason parameter munging in
RT::Interface::Web ensures that invalid byte sequences (including for
invalid codepoints, like \x{FDD0}) are replaced using PERLQQ, there are
no such guards for character strings passed to ->AddCustomFieldValue
directly via the API.

Ensure that the LargeContent passed to _EncodeLOB, upgraded from
Content, contains bytes and not characters.