This PR applies various improvements to the current setup:

• we now have two queues: the default queue has a 15 second timeout and is used for health check jobs; the updates queue has has 10min timeout
• both jobs types now implement the "unique" interface, making sure only one job instance per site is executed at the same time
• the http client uses sensible values for connect and execution timeouts to prevent slow tasks from locking up update server resources
• the rate limiting now takes various factors (site host, site ip, request ip) into account