[Bug] `DataFrame.toJson()` value column detection drops all other columns

[Jolanrensen]

Based on #2037 (comment)

Turning a DataFrame into JSON ignores all columns that are not value- or array columns if they exist:

If we read

DataFrame.readJsonStr( """[1,{"label":"record"},null]""")

we get:

     String?   Int?
        ----   ----
  0     null      1
  1   record   null
  2     null   null

and then .toJson():

[1,null,null]

label: "record" has been lost.

Expected behavior

DataFrame.readJsonStr( """[1,{"label":"record"},null]""").toJson()

should return values from (correctly recognized) array and value columns back into the JSON array, like:

[1,{"label":"record"},null]

(Related: #2046 our array column detection is completely broken because it checks for null values, however, array columns are created with emptyList() instead.)