ENH: Remote data polling and added a load abstract class

[cpelley]

• Added Pickle load and save (useful for testing/experimentation/demonstration).
• Added json load (to accompany the json save we use already use in our testing).
• 'Load' abstract class added in order to capture common capability to load UI (staging files etc.). This means EPP Load plugin doesn't have to own the logic around staging files.
• Fixed yet another issue with the logging tests (mocked the client logging so as to avoid spurious entries in parallel pytest runs).

Issues

• https://github.com/MetOffice/improver_suite/issues/2253
• https://github.com/MetOffice/epp_workflows/issues/59

[cpelley]

In an ideal network world connections always connect. In reality there are connection attempt failures and connections get dropped. I wondered if this should be a short (maybe 2 or 3 attempts) loop that will retry in the event of a failure. This could greatly improve reliability of doing the scp command.

@mo-robert-purvis, indeed. Thanks for drawing attention to this. I'm hopping that this and other potentially momentary failures (e.g. dealing with lustre file system issues) are handled by fault handling of the scheduler, but will have to wait and see how/what it can do. Perhaps adding fault handling internally will be necessary - TBD.