System deadlock when all Apache workers are busy

When JIRA is starting up it needs to communicate with Crowd, which it currently does via Apache.

If all of Apache's workers are occupied with pending requests that are waiting for JIRA to come up, JIRA cannot talk to Crowd and cannot finish starting up. This results in a system-wide deadlock.

For this kind of critical inter-app communication, we need it to happen in such a way that it cannot be blocked, even when Apache's workers are all occupied with pending user requests.