Fix #260: break import cycles in datastore w/o factories.

[tseaver]

Finishes fix for #260 for datastore (and enforces cycle check in pylint).

Alternative fix to the 'factories' indirection in #268.

Note that this requires adding a messy arg to the 'Entity.from_protobuf'
classmethod, in order to avoid the need to import '_helpers'.

Another way to avoid this cycle would be to make 'Entity.from_protobuf' into a
free function in another module (likely also moving 'Key.from_protobuf' for
the sake of consistency).

[coveralls]

Coverage remained the same when pulling edac223 on tseaver:260-break_datastore_cycles_wo_factories into 1e60e64 on GoogleCloudPlatform:master.

[dhermes]

This makes me much more comfortable than #268 :)

I think one could argue both sides for putting the from_protobuf methods in another module, but I think the benefits outweigh the costs.

The only cost is the loss of convenience of not having them as classmethods on the object.

The benefit is no cyclic dependencies and simpler to maintain code, i.e. no need to pass around a _get_value function which should be unique anyhow.

I think the loss of convenience as classmethods is not so big a deal because

• End users won't likely care about a from_protobuf method. Many end users may not even know that proto is the underlying serialization format.
• Those that do know will be able to find these methods in another module.

[tseaver]

I thought about moving the 'from_protobuf' implementations to '_helpers' (where all the related marshalling code lives), but that module is explicitly non-API. We might have to rename it and make at least those two functions APIs.

[dhermes]

SGTM with a rename _helpers -> helpers. I suppose that was bound to happen anyhow.