In a recent CI failure investigation, the server crashed following a failure to allocate a thread due to hitting a resource limit. mathias@mongodb.com has suggested that this behavior is expected to ensure that failovers happen eagerly; we wouldn't want a primary to limp along, preventing a healthier node from being elected.

Since this appears to be the reason for marking stdx::thread with noexcept, we should clarify that reasoning in the docstring. We may also want to consider examining places where we catch exceptions resulting from thread allocation attempts and remove or otherwise adapt the exception-handling logic to clarify that our expectation is for the server to crash in those instances.