[FEA] Provide option to read fields with mixed types

[GregoryKimball]

Is your feature request related to a problem? Please describe.
The cudf_experimental JSON reader does not faithfully read fields with mixed nesting.

If the field contains scalar and list then the scalar values are set to null:

>>> json_str = '[{"a":1},{"a":[1]}]'
>>> cudf.read_json(json_str, engine='cudf')
      a
0  None
1   [1]

If the field contains scalar and struct then the scalar values are set to null:

json_str = '[{"a":1},{"a":{}}]'
>>> cudf.read_json(json_str, engine='cudf')
      a
0  None
1    {}

If the field contains different levels of nested lists, the shallower lists are set to null:

>>> json_str = '[{"a":[1]},{"a":[[1]]}]'
>>> cudf.read_json(json_str, engine='cudf')
        a
0  [None]
1   [[1]]

If the field contains list and struct types, we throw:

>>> json_str = '[{"a":[1]},{"a":{"b":1}}]'
>>> cudf.read_json(json_str, engine='cudf')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/conda/lib/python3.10/site-packages/cudf/io/json.py", line 111, in read_json
    df = libjson.read_json(
  File "json.pyx", line 50, in cudf._lib.json.read_json
  File "json.pyx", line 138, in cudf._lib.json.read_json
RuntimeError: CUDF failure at: /opt/conda/conda-bld/work/cpp/src/io/json/json_column.cu:576: A mix of lists and structs within the same column is not supported

Describe the solution you'd like
We've talked about a few solutions:

• model the column as a struct that represents a union column
• add a union type to libcudf
• read the field as multiple columns and put a subscript in the name
• coerce the field's values to strings instead of trying to represent the mixed nesting

Describe alternatives you've considered
Without a workaround, JSON inputs including a field with mixed types will not be readable.

Additional context
This item still needs some design work. Please comment and share a data sample if it is impacting you.

[isVoid]

It is common in GeoJson to have different level of nested list data in the same column. Here's a column of a polygon and a linestring. (You can visualize this in https://geojson.io/)
map.zip

[GregoryKimball]

After some internal discussion we are favoring the option:

model the column as a struct that represents a union column

Here is an example of what this representation could look like:

[
  { "a": "foo" },
  { "a": { "b": "bar" } }
]

>>> df
                                             a
0         {'0': 'foo', '1': None, 'offset': 0}
1  {'0': None, '1': {'b': 'bar'}, 'offset': 1}
>>> df['a'].dtype
StructDtype({'0': dtype('O'), '1': StructDtype({'b': dtype('O')}), 'offset': dtype('int64')})

[pdufour]

This would be great :) Also running into this with this file:

(value gets set as None for first two times, then keeps the empty object)

{"Type":"insert","Key":[1],"SeqNo":1,"Timestamp":1,"Fields":[{"Name":"added_id","Value":1},{"Name":"row_key","Value":"//1=="},{"Name":"column_key","Value":"2"},{"Name":"ref_key","Value":1},{"Name":"updated","Value":"2020-12-08T02:15:07.664543Z"},{"Name":"body","Value":{}}]}

(Formatted JSON):

{
    "Type": "insert",
    "Key": [
        1
    ],
    "SeqNo": 1,
    "Timestamp": 1,
    "Fields": [
        {
            "Name": "added_id",
            "Value": 1
        },
        {
            "Name": "row_key",
            "Value": "//1=="
        },
        {
            "Name": "column_key",
            "Value": "2"
        },
        {
            "Name": "ref_key",
            "Value": 1
        },
        {
            "Name": "updated",
            "Value": "2020-12-08T02:15:07.664543Z"
        },
        {
            "Name": "body",
            "Value": {}
        }
    ]
}

[GregoryKimball]

Also see #14239

[GregoryKimball]

Closed by #14572