Refactor keys to avoid duplicate string literals

We're currently using a lot of string literals for building Redis keys, and it's relatively easy to make a typo that might go unnoticed.

It would be nice if we could improve on this.

Maybe use a separate method for generating each key?
Or perhaps simply use constants instead of literals such as `"task"` or `"executions"`, the way we already do for QUEUED, SCHEDULED, etc.?