Adding inference with JIT Scripted Torch model

[rickybalin]

Proposed changes to add inference with JIT Scripted Torch model:

• config option torch_jit determined whether model is traced or not before performing inference
• Added method torch_trainer.jit_trace_model() to jit.script the model and apply optimizations
• Bypassed Horovod and DDP initialization if performing inference since not needed

Let me know what you think!