Can't resume training using CPU

[ScottMcMichael]

The code currently does not load models within a devices/strategy scope which makes them die when using only the CPU. This may work with non .h5 model files but those don't load properly for some reason. TF 2.2 is supposed to handle model loading better within device/strategy scope so this may have to wait until we can upgrade to that version of TF.

[ScottMcMichael]

This seems to be fixed with TF 2.2 which works with the current version of Delta.