Add get_parameter_data aka get_columns

[Dominik-Vogel]

This pr add a function to the sqlite_base module and also a method to the DataSet, that allows to retrieve data in a similar way as get_data does, but instead of returning a list of rows it provides a much more useful dictionary whose keys are the parameter names and whose entries are the columns of the respective parameter as numpy arrays.

I changed the name to get_parameter_data in order to not use data storage representation specific terms such as 'column'.

This PR allows for an easy implementation of other export protocols.

It furthermore establishes the basis for a fast compiled implementation of the same.

[codecov]

Codecov Report

Merging #1400 into master will increase coverage by 0.16%.
The diff coverage is 97.7%.

@@            Coverage Diff             @@
##           master    #1400      +/-   ##
==========================================
+ Coverage   74.09%   74.26%   +0.16%     
==========================================
  Files          85       85              
  Lines        9883     9946      +63     
==========================================
+ Hits         7323     7386      +63     
  Misses       2560     2560

[jenshnielsen]

I will test this out further on some examples but I think this looks good

[astafan8]

It looks good! Just needs to be tested for all the paramtypes, I guess.

Question: do I understand correctly that it is the sqlite_base.get_parameter_data function that is planned to be completely substituted with the compiled version? Not the sqlite_base.get_data?

Question: do I understand correctly that this PR intentionally ignores the fact that get_parameter_data is NOT faster than get_data?

[Dominik-Vogel]

@astafan8 thank you!

1. It will not be completely substituted, but it will rather form the entry point. Calling it results in looking up the presence of the compiled code. The query will still be build as is and then instead of executing it on python it will be executed from the compiled library. No,get_data will not be substituted. This is because getting the data in form of a dictionary of columns seems a lot more natural.
2. Yes, it does ;-) As far as I can see there is no way in python to implement such a query faster. I have benchmarked some ideas but could not get any improvement.

[jenshnielsen]

I did a bunch of manual tests with combinations of array type and regular parameters. They seem to work great. I will transform them into regular tests. @Dominik-Vogel is it ok to push here?

[Dominik-Vogel]

yes feel free @jenshnielsen !

[jenshnielsen]

@astafan8 I think this addresses all the issues you found

[astafan8]

Thanks for taking the effort of enhancing the tests and docstrings! It's also good to see that adding more tests didn't require significant changes in the implementation.