Padim Training with large dataset

Application gets out of memory while training, figure out a way to split training in stages, so application can be run on machines with moderate amount of memory resources. HDF5 filesystem could be solution to this.
Functions:

- ``extractEmbeddingVector``
- ``extractEmbeddingVectorsDataloader``
- ``getMeanCov``