Allow implementors to define field order

[jrdh]

This comes off the back of NaturalHistoryMuseum/ckanpackager#47.

When importing data into the datastore we store the field order from the source, however, if we don't have any of these details (say, for example, the data was sideloaded directly into mongo and elaticsearch) then we just use the fields from the resource's index mapping in elasticsearch. These fields are ordered alphabetically to maintain some level of consistency.

It'd be nice to allow implementors of the IVersionedDatastore interface to reorder the fields, perhaps:

def datastore_modify_fields(self, resource_id, version, mapping, fields):
    return fields

This allows the implementor to both override the field order and change the fields if they want.