Those are types for which the Sprintf("%+v") catchall code is used.

The output of Sprintf("%+v") is neither readable nor parseable. JSON encoding
is better in both regards.

There are some downsides:
- Unexported fields don't get encoded anymore. However, the same happens
  with other backends (in particular, zapr), so one could also
  argue that making the text format behave like those others
  is good because developers notice early that they custom
  String and MarshalLog methods if they want to log such fields.
- The result KObjs (a slice) now gets encoded as array of structs.
  KObjs already got deprecated a year ago. KObjSlice should be used
  instead.