Related to
#444

The rationale behind this is that we don't want to pollute worker job
queue with large file blobs. Instead, upon API request, we create a
pg_largeobject and use that in worker. After job is finished (with
error or not), the object is removed.

Related to
#433

Textual formats often have either encoding or column errors.

This is a common issue, see e.g.
https://duckdb.org/docs/data/csv/reading_faulty_csv_files.html

We just ignore them.

Also, some small refactorings.

Related to
#418

These types are the same, except for Database.Empty

I managed to have backwards compatibility with the frontend format,
hence the frontend doesn't need any changes.

This simplifies the WithQuery structure even more

Also, add WithQuery pubmed test (with api_key)

There is now only a 'datafield' field, no need for duplicated
'database'.

Related to #441

[graphql] fix droping field prefixes

See merge request !379

It wasn't used anyways.

I want to understand ngrams algorithms better.

Upgrade 'haskell-bee' version, also bump up versions of some packages

See merge request !377