global: improve exception handling

[mvidalgarcia]

Current behavior

The exception handling of this workflow engine is different from Yadage an Serial ones, where exceptions are propagated to the reana-commons's workflow engine command wrapper. This wrapper takes care of setting the workflow status as failed in case of failure, as well as propagating the logs to the client.

In case of CWL, exceptions are handled locally (See [1], [2], [3], etc.), and eventually propagated to internal cwltool functions.

⚠️ An affected feature of this issue is the k8s user-custom memory limits per job. By using this feature, if the user provides a wrong memory limit (wrong format or over the limits), the exception that occurs is not properly caught and causes the workflow engine to stay forever in running state. More info here.

Expected behavior

Exception handling should behave as Serial and Yadage, where job submission exceptions are caught by reana-commons.