Currently we wait to run the Windows tests for a commit until all the Linux tests have passed. This results in unnecessarily long feedback time to know if a commit passed on Windows.
The only benefit of staggering is a small efficiency in resource usage when a breaking commit is merged to develop, but optimizing for failure seems unproductive when our intent is to maintain a healthy CI pipeline. On the other hand if we want to embrace staggering to optimize resource usage on failure, then we should also split up JDK8 and JDK11 into two stages rather than run them concurrently...
