recreate quantized shards of face recognition model, which apparently fixes the issue with karma loading the binary files + include quantized model tests for face recognition model
