Grist has needed a job queue for some time. This adds one, using
BullMQ. BullMQ however requires Redis, meaning we couldn't use
jobs for the large subset of Grist that needs to be runnable without
Redis (e.g. for use on desktop, or on simple self-hosted sites).
So simple immediate, delayed, and repeated jobs are supported also
in a crude single-process form when Redis is not available.
This code isn't ready for actual use since an important issue
remains to be worked out, specifically how to handle draining
the queue during deployments to avoid mixing versions (or - if
allowing mixed versions - thinking through any extra support needed
for the developer to avoid introducing hard-to-test code paths).
