Scope

Currently when exporting a collection to CSV, the headings of the file were deducted from the first batch of the query result. In case the user choose to include relational data in the export and the first batch of items did not include any relational data, the headings did not exist and therefore the json2scv library put the values in the wrong 'column'.

In this PR, the field names which should be exported and hence the headings of the csv file, are determined beforehand instead of deducting those from the first batch.

An example

• setup 13 articles, first 10 don't have an author related to it
• set EXPORT_BATCH_SIZE of 10
• specify those fields: id, title, author, author.first_name

Before

For the first batch, the author is always null. When deducting the headings this way, no first_name column is present in the export. For the second batch though, this value existed in the database response and was added in the CSV file in the wrong place. within the author column there should be the ID, and "Jan" should be in a column which doesn't exit.

After

With the fields specified explicitly, the columns of the export did not rely on the first batch result anymore. This way the shifting could be eliminated.

NOTE: I could not test it with the dataset provided in the linked issue, since this is a dump from v10.

Fixes #22298