Generate debug output files for all regression tests, even if they're not checked

[alexdewar]

We only check the contents of the debug output files for the regression test for the simple example (to avoid unnecessary churn associated with regenerating them all the time), but they are useful for debugging (hence the name!) so it is probably a good idea to generate these files for every test, even if they're not checked. As the output files are uploaded to GitHub as test artifacts, having this extra info could be useful for debugging failing tests (see e.g. #1185).

We should probably ensure that we are still left with at least one integration test with the debug_model setting disabled though (maybe we should add another for simple?).

[AdrianDAlessandro]

I'm not entirely sure what you want from this. I can interpret it multiple ways. Which of these options is preferable:

• Keep all tests the same but include the debug files in the output directory (i.e. generate them but don't test them). This should not require changing any of the tests/data files.
• Make all the tests include debug files but add an extra one that doesn't. This will require adding the debug output files to the tests/data

[alexdewar]

I'm not entirely sure what you want from this. I can interpret it multiple ways. Which of these options is preferable:

• Keep all tests the same but include the debug files in the output directory (i.e. generate them but don't test them). This should not require changing any of the tests/data files.

This one ☝️. We used to include all the debug files for all models, but it created a lot of churn and we agreed we didn't need to test every file for every model, so dropped them. It would be good to have these files included with the test artifacts though, so e.g. you can get more information about why a test might be failing on macOS, even if you don't have a mac.

• Make all the tests include debug files but add an extra one that doesn't. This will require adding the debug output files to the tests/data