Check the caching setting with python path need to be valid

[unkcpz]

fixes #6064

The identifier is checked when calling verdi config set -a caching.enabled_for, it is not raised in the first place because the validator only checks the format of the identifier but does not actually validate if it is a loadable module/class.
Look at other tests which are using identifier = "some_ident", those are now all failed.
I think it is worth to use importlib to check the idendifier string more strictly, what do you think? @sphuber

[sphuber]

Thanks @unkcpz . This obviously changes (breaks) current behavior as it will now break existing caching configurations that might include rules that do not match importable resources. Also, it is a bit asymmetric, as the entry point based rules are not checked whether they can be loaded.

So here is a suggestions for two changes:

1. Add the same check for entry points, trying to load it and raise if it cannot be loaded
2. Make the import and entry point load-check optional by adding a boolean flag strict which is False by default

For the latter, we can decide whether we want to make it True by default in some future release that will allow backwards-incompatibility.

[sphuber]

@unkcpz I implemented the changes I suggested. Please have a look. If you think the changes are good, I think we can merge this.

[sphuber]

I might just need to update the docs a bit.

[unkcpz]

@sphuber Really sorry that I didn't manage to update and finish this, and thanks a lot for taking care of it.
Yes, the changes look good to me. Cheers!