Integration test improvements and a few bugfixes

[na--]

Split among the first 3 commits are various improvements and additions to the existing integration tests:

• proper exit code detection in integration tests
• mocking of the cloud API and checks for run_status and result_status values in k6 run --out cloud tests
• support for using and testing the k6 REST API in integration tests, without port collisions
• support for mocking signals (like Ctrl+C) mid-test run

The 4th commit actually fixes #2788, a small bug I noticed while trying to remove the whole core.Engine (#1889). However, because the Engine removal will likely be merged at the start of the v0.43.0 cycle and because the bug was easy to fix even in the current implementation, I decided to fix it now.

This way we'll get the fix released in k6 v0.42.0 and I'll have the test already present in master, in preparation for the bigger PR.

The 5th commit fixes a minor data-race with the progressbars and masks a few others with dangling goroutines after errors in the init phase, see #2789 (comment) for details. And the 7th commit fixes another minor data race in the old Engine, see #2789 (comment).

[na--]

I added one more commit to fix a data race and mask a few others, where goroutines spawned by go run persist and log things after cmdRun.run() and even newRootCommand().execute() have finished executing... 😞

As I wrote in the FIXME and TODO comments, the proper fix will have to wait until #2790 and #1889 have been implemented, hopefully in the next cycle.

[oleiade]

I left a couple of non-blocking nitpicks. Really great work, glad to see those improvements 👏🏻

[imiric]

Nice work! 👏 Just left some minor comments.

[na--]

These tests were more useful than expected, found another data race in the current Engine... 😭 The Engine didn't wait for the run_status to be properly propagated to the outputs before it finished executing, leading to a potential data race.

This will also be fixed in the new architecture, it's just one more reason why the current Engine architecture sucks. Still, the fix was relatively easy (129fdb6), so I chose to do it rather than disable the test that triggered it.