Invalid workflows fail post-detach

[oliver-sanders]

When [scheduler][run hosts] are used to run scheduler on, validation errors occur post-detatch which makes the failure invisible.

E.G. when you run an invalid workflow:

$ cylc play <id>
...
<id>: <server> PID=<pid>

It looks like it worked, but it didn't:

$ cylc cat-log <id>
...
ERROR - Workflow shutting down - WorkflowConfigError: This workflow requires an initial cycle point.

This didn't use to be the case, the change was likely caused by recent startup reordering.

Note: #3896 will reduce the pressure on this, however, ideally we would fix this in the cylc play command properly.

Pull requests welcome!

[MetRonnie]

I have reproduced this even without [scheduler][run hosts], so maybe higher priority?

#4836 is the culprit, however I'm struggling to figure out why just by looking at the code changes. Will try more in-depth debugging...