Implement prefetching for ClusterTaggableManager

[chosak]

This commit implements prefetching for ClusterTaggableManager so that prefetch_related works properly:

class TaggedPlace(TaggedItemBase):
    content_object = ParentalKey(
        "Place",
        related_name="tagged_items",
        on_delete=models.CASCADE,
    )

class Place(ClusterableModel):
    tags = ClusterTaggableManager(
        through=TaggedPlace,
        blank=True,
    )

places = Place.objects.prefetch_related("tags").all()
for place in places:
    place.tags.all()

Currently the above code functions properly but does not actually leverage the prefetching; each call to place.tags.all() re-queries the database for each place's tags, instead of using the list previously fetched with the prefetch_related("tags").

This commit properly implements prefetch_related to avoid those duplicate queries.

To test, run tox.

This change requires a bit of complexity due to a change in Django 5.0+ (which is tested under tox but not currently tested in GitHub Actions): As documented in https://code.djangoproject.com/ticket/31486, it's no longer possible to pass unsaved model instances to related filters, preventing calls like:

TaggedPlace.objects.filter(content_object=Place())

This requires some special handling to continue to support querying tags on unsaved model instances:

place = Place()
place.tags.count()

Please see comments in taggit.py for additional detail.

Closes #38, and may also address wagtail/wagtail#6044.

[gasman]

All looks good to me. Thanks for the detailed comments!