Make Sidekiq routing rule validation noisier

Extracted from #1175 (comment 627102302)

Recently, if a user feed an invalid routing rule to the configuration file, Sidekiq client simply puts a message into the Application error logs, and fallbacks all jobs to per-worker queue. The code locates here.

This guarding behavior is not ideal from multiple perspectives. As @cmiskell pointed out, it creates more issues in long-run:

As an on-call SRE, both are terrifying, but a crash is actually better in a kubernetes world because the deployment will fail and get everyone's attention, without breaking the site; new pods that can't start up will retry in a CrashBackoffLoop and helm will eventually notice/give up, without spinning down the currently functional pods. Then someone will come along and look at why the deploy failed and notice some logs (hopefully). So for us (GitLab), that's fine. It sucks a bit on VMs (for customers for example), but hopefully they're paying more/different type of attention.

Workers being sent to the wrong queue would be utterly disastrous for us; our shard sizings are based on known workloads, and if everything ended up on catchall (or some other shard) we would have a massive outage to deal with (worse even than recent days).

Therefore, we should make this validation noisier, maybe just let the application crashes.