We currently run the tests in parallel as that's the default on Cargo. To ensure certain tests don't overlap with each other, we use the test lock, which can be acquired exclusively or with other concurrent tests.

This generally works, but if we ever forget to acquire the lock, it can cause hard to debug failures in other tests. Also, the test output can be hard to read since it's all intermingled. Lastly, the timings for individual tests can be incorrect, since a majority of the time spent for many of the tests is waiting to acquire the test lock.

We should consider running all our tests sequentially to alleviate these issues. First, we'll need to verify that doing so doesn't increase the build times by too much.