deterministic behavior for elements that are union of type T and list of type T

[rudolfix]

Often json document contain inconsistently typed elements that may occur 1 or more times. In case of 1 they are inserted directly as type T, if > 1 they are inserted as list of type T.

ie. first_name may be "Juan" or ["Juan", "Pablo"]

in that case DLT builds bot flattened representation and table representation for T and inserts in both place.

Fix the json normalizer such that if table is present in the schema, the table is always used.
in relational.py

if isinstance(v, (dict, list)):
                if not _is_complex_type(schema, table, child_name, __r_lvl):
                    # TODO: if schema contains table {table}__{child_name} then convert v into single element list
                    if isinstance(v, dict):
                        # flatten the dict more
                        norm_row_dicts(v, __r_lvl + 1, parent_name=child_name)
                    else:
                        # pass the list to out_rec_list
                        out_rec_list[child_name] = v

Note: in case of automatic schema inference, all values until first list of T is detected will be added flattened. To assure consistency such tables should be added to the schema before first data item is normalized