A struct field can be referred to as "RECORD" or "STRUCT", and this commit assures that the to_api_repr() logic is correct. It also moves STRUCT_TYPES from _pandas_helpers to schema where it belongs more naturally IMO.

Discovered this while writing the "all types" system test, and tried to create a STRUCT field using the standard SQL type name "STRUCT".

How to test

• Using the BigQuery client, try to create a table with a "STRUCT" column (using that particular type name):

  from google.cloud import bigquery
  
  bq_client = bigquery.Client()
  
  schema = [
      bigquery.SchemaField(
          "person_struct_field",
          "STRUCT",
          fields=(
              bigquery.SchemaField("name", "STRING"),
              bigquery.SchemaField("age", "INT64"),
          )
      ),
  ]
  bq_table = bigquery.table.Table(
      table_ref="{}.{}.active_users".format(PROJECT_ID, DATASET_ID),
      schema=schema,
  )
  
  bq_client.create_table(bq_table)

Actual result (before the fix):
The server responds with an error:

Field <field_name> is type RECORD but has no schema

Expected result (after the fix):
The table is successfully created.