fix(serializer): propertyFilter should apply to arrays as well

[nikophil]

Q	A
Branch?	3.1
License	MIT

PropertyFilter does not presently work with arrays of objects which are not ApiResource: it just always serializes these objects as null arrays.

FYI I've tested with ArrayCollection of objects which are not ApiResource and it works like a charm

[soyuka]

todo: check if we shouldn't remove operation also

[nikophil]

could you explain me what choices lead to unset() these contexts?
because we're serializing "non api resource" objects, there would be no need to have the operation available in the context?

[soyuka]

because it'll re-use the old operation if any when creating an IRI

[nikophil]

unless I'm misunderstanding something, no IRI should be generated, since we're dealing with non-apiresource objects?

[soyuka]

Yes they have Skolem IRIs unless disabled by using the genId: false attribute.

[nikophil]

🎉

[temp]

This MR broke our code in unexpected ways, we now receive "Unable to generate an IRI for the item of type X" in our unit tests.
Our resource "Wheelset" has a property of type array, and Wheel has a property of type WheelManufacturer, which is a resource itself.
The wheel-array is now normalized within the if-block of this MR in AbstractItemNormalizer, and leads deep down in IriConverter and IdentifiersExtractor to the above error. We haven't got a clue what to change. When we comment out the new if-block, everything works as expected. Seems like with the new if-block it tried to generate an IRI, without it it doesn't try to generate an IRI for the nested resource.