The custom text fields didn't have the same behavior that standard model

fields had. Switching text types without setting new text didn't result

in automatic escaping/unescaping logic, and the proper fields weren't

being read consistently everywhere. The fields also didn't respect the

deprecated text_type= property anymore (whereas they did in previous

releases, before the split fields).
We now have logic that mirrors what we do for plain text fields, and

helps normalize the results so that there's always an entry for the text

and an entry for the text type.
Unit tests were updated to check custom fields when setting text types.

Unit tests pass.
Tested that I could change the text type field in API requests and the text

would auto-escape/unescape.
Tested having the text set to Markdown in extra_data, and just sending new

text in the request. Saw it was auto-escaped.
Tested custom fields in the UI.