Can we add integration tests?

[lostmsu]

Does not seem like its urgent, but running a time-limited analysis on something as big as scikit-learn (see #379) before merging pull requests would alleviate some unnoticed regressions.

[jakebailey]

You're describing two things:

• Running automated tests on PRs and such, which is already covered by Set up CI with tests on Windows, Mac and Linux #352, but may need to wait until we simplify things to the point where running the suite doesn't take half an hour.
• Running tests on big complicated codebases like scikit-learn, which we might not really be able to do either until we remove the slow analyzer either (from experience running it on scikit-learn takes a good 20 minutes even when I do it from /tmp).

[lostmsu]

@jakebailey I think tests can still be run on scikit-learn, or any other large project as long as they are time- or iteration-limited. E.g. just have AnalyzeQueuedEntries called with a timed cancellation token, or cancel it as soon as, say, 100000 units have been (re-)analyzed. The later option would be more stable.

[MikhailArkhipov]

To #352