GitLab

The functionality provided by the zoe_client package no longer fits a single ZoeClient class.

The scheduler only knows about executions. Each execution has an application ID that can be used by the scheduler to group executions coming form the same application. The client has its own database that manages users and applications.

Split configuration into different modules for the scheduler and the clients

Create a scheduler policy base class and make the simple scheduler policy inherit from it.
Document all methods for the scheduler and policy classes and make sure Sphinx generates the documentation correctly
Pass the full application description to the scheduler policy
Smaller cleanups

 This commit also removes completely the reverse proxy code, as the same result can be achieved with a socks proxy and much less complexity in Zoe's code.
  Everything that is extraneous from the scheduler is being moved away.
  In this commit zoe_web is broken.

To appease the pypi gods who dislike markdown.

Move main() script code inside packages to satisfy the best practices for python packaging.

Remove module-level initializations so that test cases can be built more easily.
Add one test-test case for the object_storage module, just to make sure the boilerplate is right.

It is now easier to write tests, also checked where sessions are created and have them correctly closed at the end.

Closes #6

Conflicts:
	README.md

Add new server and client classes
Convert user ops and platform statistics to use new RPC

Refactor the state classes to differentiate between State classes, coming from SqlAlchemy, "normal" classes that are moved areoung and passed to client applications and statistic classes that are generated when required, on the fly.