The old method (using dict() and _fixup_dict) create clones from IFD_v1,
but it read all items when iterating, losing all benefits from the lazyness
of IFD.
The new method use a custom subclass that try to prevent all iterations
as much as possible, keeping all tags undecoded.

The return value isn't a dict anymore, but a subclass of ImageFileDirectory_v1
that doesn't support tov2(). All decoded tags are formatted as before.

The _fixup does exactly the opposite

Co-Authored-By: Glandos <bugs-github@antipoul.fr>

Co-Authored-By: Glandos <bugs-github@antipoul.fr>

Co-Authored-By: Glandos <bugs-github@antipoul.fr>