Create an end-to-end test

[igelstorm]

What

• Sets up Maven Failsafe, the standard way to run integration tests in Maven
• Translates the run-simpaths GitHub Action into an integration test (which does the same thing)
• Adds assertions about the model output (the Statistics*.csv files)
• Changes the GitHub Action to run these tests instead of directly running the model

Why

Addresses #138. This will allow major changes to implementation details (e.g. optimisations, design changes) to be done much more safely, since these tests passing gives some assurance that the overall behaviour of the model hasn't changed.

How to use

• The new integration tests can be run by running mvn verify.
• Since several steps are time-consuming, I've split them into multiple tests. It's possible to run an individual test by running e.g. mvn verify -D it.test=RunSimPathsIT#testVerifySimulationOutput

[andrewbaxter439]

This all looks pretty awesome and very neat! Nice to be able to have end-to-end all handled within maven/JUnit.

Have suggested changes but actually not sure they're implementable. If parallelising three types of tests doesn't work then can just delete other tests.

[andrewbaxter439]

One small, additional though - is it worth doing/possible to force the simpaths setup to use only the training data, for speed and consistency when running mvn verify? At the moment it chooses based on the presence or absence of population data outside the training folder in 'Initial Populations':

https://github.com/centreformicrosimulation/SimPaths/blob/e1e3c79b65bc704053898b3061a557f693e669c5/src/main/java/simpaths/experiment/SimPathsStart.java#L207-L213

Think only way of overriding this would be to pass another command-line argument which is in turned passed to runGUIlessSetup to then override the setting of the training flag. If that's a different job though then can leave out of here.

[igelstorm]

One small, additional though - is it worth doing/possible to force the simpaths setup to use only the training data, for speed and consistency when running mvn verify?

This is a good point - it would definitely be good to make it easier to run locally even if you have other initial population data installed (without having to delete/move files).

Ultimately I think the ideal is if it's always possible to configure which folder is used for input data; then this test could just make use of that feature. I'd be reluctant to add a less thought-out, potentially more awkward fix purely for use in the tests, though - so I suggest we deal with this separately.

[andrewbaxter439]

This all looks pretty smashing! Hope this 'resolves' my prior review comments