App Runner B-Fabric integration

[leoschwarz]

• separate the submitter related parameters out of the raw_parameters into a dedicated WorkunitDefintition field
• workunit.mk should set the correct bfabric instance into an environment variable
• probably redundant call (workunit save status processing)
• check if we can organize the registered data a bit better (actual executable rather than yaml, and wrapper creator <-> submitter split)` i merged the two components

Essentially, this PR implements a new submitter/wrapper creator pair for bfabric-app-runner conformant apps. Basically, it makes it super easy to execute any app which is defined by an app_definition.yml into B-Fabric.

The main benefits are the following:

• Full support of slurm configuration, at submitter, application, and workunit level. -> This will allow us to use our resources more efficiently.
• Configuration is almost fully contained in a YAML file which allows to deploy this for the test system without conflicts. We even "fake" the output staging to a SSH Host and directory, without necessitating the configuration change in B-Fabric test. -> We can run our apps on the bfabric-test instance now.
• Workunits can automatically be linked with a viewer for live log updates. -> This will simplify the logging in the future, because there is only one log source that needs to be handled.
• Application folders in /scratch are automatically created from the configuration only. -> This is a further step towards more ephemeral scratch directories which will allow to scale our data processing more.

Caveats:

• This new functionality is not available for legacy apps.
  • At least testing the legacy apps on the test system will be problematic, since they manage their input/output staging and scratch folder allocation and thus it cannot be avoided that they will write to the production stores easily.
  • We could implement a compatibility layer with the above caveat, but except for avoiding a duplication of efforts I don't really see much to be gained.
• We do not create resource entities, for the logs directly, but rather links.
• Some specifics are still embedded in the shell script which is autogenerated, though it is arguably a lot more generic than it was before.