Continuous benchmarking with examples

[NOBLES5E]

• [x] Provide a script running all examples like https://github.com/pytorch/examples/blob/master/run_python_examples.sh but for multiple machines
• [x] Provide numerical results for first two epochs on each example in deterministic mode
• [x] Provide internal training job template that automatically benchmark and compare numerical results to ensure the newest Bagua works correctly