Worker: Better error handling for credential errors

[josephjclark]

Closes #600

This is actually a fix down in the engine.

When loading credentials from Lighting, if any credentials fail to load, we catch and report the error, and throw in the main validation function.

Previously, because of how credentials are loaded in nested promises, this case resulted in an uncaught exception and everything would blow up. It's more stable now - and we should get a nice error message to explain the fail reason!

[josephjclark]

Hey @taylordowns2000 are you able to retest against this branch?

[josephjclark]

I am unable to reproduce the original issue on my lightning.

The changes here are well worth merging IMO.

As a result of this branch, Taylor sees these logs (edited for clarity):

[RTE] ❯ Loading credential a671b76f-8173-4564-b354-64695aa60240 for step 4415c14f-bb32-4407-a38a-cacf1a7fc49a
[SRV] ✘ error connecting to run:9f9c471a-79f1-4627-bd2f-17b9183ff2c1 {
  "reason": "unauthorized"
}
[RTE] ❯ Error loading credential a671b76f-8173-4564-b354-64695aa60240
[SRV] ✘ ERROR in workflow-error handler: timeout

I think that when the credential fails to load, the worker gets kicked out of the websocket at the lightning end. It is no longer allowed into the run channel.

When the credential fails to load, the worker will try to send logs and events to Lightning. But these fail because the channel is dead - and so we get the timeout on the workflow-error handler (which really means run-error, which has been fixed on this branch)