Add embed schema parameter and embed_only field option

[bpneumann]

Implements the functionality described in #151.
Similar to the functionality of only, but allows fields to be marked as embed_only. A common use case for this is for schemas which contain large collections which should only be serialized in specific cases.

Add support for embed_only option on fields, and as a Meta option in schema classes.

If a field is marked as embed_only, the field will only be serialized if it is included in embed schema parameter.

Example:

    class MySchema(Schema):
        foo = fields.Field(embed_only=True)

    sch = MySchema()
    assert "foo" not in sch.dump({"foo": "bar"})

    sch = MySchema(embed=['foo'])
    assert "foo" in sch.dump({"foo": "bar"})

This also supports nested fields using the same dotted identifiers used by the only option.

    class ChildSchema(Schema):
        bar = fields.Field(embed_only=True)
        baz = fields.Field(embed_only=True)

    class ParentSchema(Schema):
        foo = fields.Nested(ChildSchema())

    sch = ParentSchema(embed=['foo.bar'])
    dump = sch.dump({"foo": {"bar": "val", "baz": "val"}})
    assert "foo" in dump
    assert "bar" in dump["foo"]
    assert "baz" not in dump["foo"] 

[sloria]

I believe this is the same use case described in #244 . See the "include" recipe for a way to address it.

Holding off on this PR for now.

[bpneumann]

This recipe works for schemas without nested fields, but offers no way to include fields on nested fields.

Extending the example from #1737 (comment)

class Organization(BaseSchema):
    members = fields.Nested(Member())
    name = fields.String()

organization = Model(name="Book Club", members=[member])

pprint(Organization(include=("members.photo",)).dump(organization))
# raises KeyError

Since members.photo is not a field on Organization, this raises a KeyError. The desired behavior would be to set include on the Organization schema to members, and propagate photo to the include parameter of the nested Member schema.

A full solution to this can currently only be accomplished by subclassing Schema as well as Nested to accommodate the include option, and overriding the initializers, _init_fields, and _normalize_nested_options.

Alternatively, this could be made simpler with the changes proposed in #1725 allowing subclasses of schema to update fields before serialization/deserialization according to extra arguments.