Make materialisation of columns optional for the library extract query.

[piotrszul]

Currently the extract operation performs materialisation of extracted columns (that is conversion to the corresponding literals). This is a desirable behaviour for the FHIR API extract operation, when the values of for the columns should be simple values representable in CSV format, but it puts a limitation on the types of columns that can be extracted.

For the library API hover the result of extract is a dataset, in which any type can be represented, so this constraint is not necessary.

The API should allow the user to extract any columns they want and preserve the original types of these columns. (In particular currently DATETIME (TIMESTAMP) columns are converted to STRING columns).

The API should allow the user to optionally request materialisation.

Additionally the users should have an option to materialise selected columns by adding type conversion (e.g. to STRING) these columns.