Better workers management

[joelcogen]

Several things should be improved

• request a build and invite user crash if Redis is down
• when Redis is restarted, builds requested during downtime are never restarted

Either we make the dependency on Redis stronger and block everything if Redis is down (which I guess would make sense), or we rescue those errors and add some worker or something to find what need to be re-enqueued when Redis restarts