Separate glTF2 model loading from uploading it's vertex data to GPU

[IAmNotHanni]

Just as with textures, we can separate the loading of a glTF2 model using tinygltf from the actual upload of data to the GPU. This way, we can run models in separate threads while the engine's Vulkan backend is being initialized. This all falls into the regime of parallelization of the engine using taskflow and creating an asynchronous resource loading mechanism.

We need more planing + discussion about loading resources (textures, sounds..) asynchronously.

[IAmNotHanni]

This ticket turned out to become a more general idea in the engine: We should split up loading, preparing and rendering of resources. The same principle can be applied to octree rendering for example.