Add FittingResult model

[jonc125]

Probably in a fitting app, cf #133?

See https://github.com/ModellingWebLab/project_issues/wiki/Workshop-notes-2018 for ideas.

Viewing results should hopefully just reuse code from #131 etc. But we'll need to do (at least some of) #135 before this can be fully implemented and tested.

The plan is being sketched out further in #203 and #241. Possible steps:

• Do the Runnable refactor outlined in First outline of fitting result models #241, with DB migrations, ensuring tests still pass etc. This can be a single PR.
  • This will mean that the RunningExperiment table won't need changing for fitting experiments, and indeed receipt of finished experiments in experiments/processing.py:process_callback (and cancelling running experiments) won't need any additions.
• Then building on that and Use foreign keys for model/proto version references in Experiment table #250, create the FittingResult model inspired by the outline in First outline of fitting result models #241. No views initially, just tests of the model directly? (Like experiments/tests/test_models.py)
  • So create the FittingResult and FittingResultVersion models.
  • FittingResult needs to link to repocache tables for entity versions, not use SHA strings.
• Then basic views along the lines of experiments/views.py, trying to reuse code where possible as was done for the specs & datasets:
  • FittingResultVersionListView
  • FittingResultVersionView
  • FittingResultVersionJsonView
  • FittingResultDeleteView
  • FittingResultVersionDeleteView
  • FittingResultFileDownloadView
  • FittingResultVersionArchiveView
• Adapt submit_experiment to create FittingResult etc. instances, refactoring to share code where possible.
• Harder views, probably need JS code rework e.g. as discussed in Create ExperimentalDataset app with basic model & views #131
  • FittingResultComparison[Json]View

General notes:

• To get submission of fitting experiments to work we'll need a 'fitting' version of experiments/processing.py:submit_experiment that creates the FittingResult etc instances. We may be able to refactor so they share some code though.
• We don't want to try automatically migrating existing hacky fitting specs (stored as Protocol instances) to FittingSpec instances etc. Instead manually re-create ones we care about (just one really! i.e. Kylie since wave cell 5).
• We'll need to consider how best to test this as we develop, including building enough of the fitting experiment submission UI Design UI to run fitting experiments #135 to test it out manually. Perhaps better to use a sample result from the current hacky implementation as a test case? (Maybe downsample the 3MB CSV for speed!)