Refactor registrar into multiple files

[stuartmcalpine]

In preparation for us doing more with entries other than creating them, I've reformatted the registrar class into one for each table, e.g., dataset, execution and dataset_alias.

As before they all follow the same format with a slight tweak. e.g., datareg.Registrar.register_dataset() now becomes datareg.Registrar.dataset.create() and datareg.Registrar.execution.create() etc.

This is in preparation for new functions like datareg.Registrar.dataset.modify() or datareg.Registrar.dataset.delete() etc. Placeholders for these functions have been added.

It also tidies the code a bit (mainly in prep for the new functions) so that registrar.py didn't become very long.

Right now no internal or external functionality has changed (other than the name of the function to call for registering datasets).

Will update docs to reflect changes in another branch.

[JoanneBogart]

I've suggested a couple specific and simple changes, but additionally we should probably think a little more about the organization. One possibility would be to make a base class - call it RegistryTable for now - which your new classes would subclass. RegistryTable would have base implementations for at least modify and delete. register is generally more complicated and different from one table to the next, so perhaps RegistryTable.register( ) would always have to be overridden by the subclass. To the extent that the different tables need, even partially, the same treatment, the base class would be a convenient place to put the common parts of the implementations.

[stuartmcalpine]

I have updated the structure as suggested. Now there is a base table class which the execution/dataset/etc tables will inherit. The base class holds the delete and modify function as they should be universal.

The register (renamed from create) functions are in each of their own subclasses.

The only slightly dirty bit is that I have to pass the execution subclass into the dataset subclass in the Registrar class. Thats because we need to create an execution from within the dataset registration also. But I dont see a way around it.

[JoanneBogart]

It looks good to me. I made one trivial comment.