mv_machine_learning: drop Mediavision dependency from landmark detection adapters

mv_machine_learning: drop Mediavision dependency from landmark detection adapters

[Issue type] : code refactoring

Drop Mediavision dependency from adapter classies of landmark detection task
group. There is a use case that pre-trained model file is used in private.
Therefore, the model relevant code cannot be opened. And even such users want
to configure the inference or training engines in their way.

In this case, we need to manage it properly by providing plugin approach of
behavior class. And this patch is a first step for supporting plugin based
behavior class which can be delivered as separate package.

Change-Id: I54a3f2f8de86290718129b8a44a724d3d9a3f246
Signed-off-by: Inki Dae <inki.dae@samsung.com>