In Memory only Dataset

[jenshnielsen]

Implements an in memory dataset that can be used to measure without writing the raw data to the db.
Furthermore this enables you to reload a dataset from a netcdf file and then write the metadata back to a db file and a plot it / explore it in other ways

• Write a changelog entry

[codecov]

Codecov Report

Merging #3094 (fc38c14) into master (de6e7de) will increase coverage by 0.38%.
The diff coverage is 90.60%.

@@            Coverage Diff             @@
##           master    #3094      +/-   ##
==========================================
+ Coverage   65.31%   65.69%   +0.38%     
==========================================
  Files         223      225       +2     
  Lines       29969    30463     +494     
==========================================
+ Hits        19573    20012     +439     
- Misses      10396    10451      +55     

[jenshnielsen]

Things to figure out.

• If a dataset is loaded from netcdf should we add it to the runs table in the current db. No but see below.
• Need to handle attempt to write extra metadata for a dataset loaded from netcdf to a file that does not have the dataset
• Should there be a function to add a dataset including the results table back to the db (yes but done outside this pr)
• Currently changes to metadata of any kind will not be written to the db except when the run is marked started and completed. It will also not be added to exported data after the data has been exported. This is now done
• How do we handle the __len__ attribute. This is somewhat sqlite specific since it counts the number of sqlite rows. Captures the number of points written.
• Upgrade create_runs to somehow mark the runs table for not containing the data but it being external
• Not strictly here but important. Add test for complex data export. Handled by Export complex numbers to netcdf #3126
• Test that data including complex can round trip correctly when exported to netcdf