GoodJob.restart should not start capsules (job execution) when in a webserver but not in async mode

[bensheldon]

Fixes #1054

This isn't my favorite, but the docs have been telling people forever to use this pattern and there isn't a simple way to fix #1054 while also not requiring everyone to change their configuration.

# config/puma.rg
before_fork do
  GoodJob.shutdown
end

on_worker_boot do
  GoodJob.restart
end

If this causes problems, I imagine that I'll add something like restart! that will ignore the new check (or introduce a simpler "start", or ensure that "restart" doesn't actually start the capsule if it hadn't previously been running but rather just return it to the same state). But I am not aware of anyone manually creating capsules, so it's likely not used.