save validation loss history in a csv

[msrepo]

Is your feature request related to a problem? Please describe.
train.py has records of validation loss history. We need to persist this history in to a file.

Describe the solution you'd like
save both training loss and validation loss every epoch. Saving training loss helps find if we are overfitting/underfitting.

Describe alternatives you've considered
we have manually kept a list of metrics each epoch. we could just save it into csv. Also, need to be careful we write the csv every epoch in case the training gets terminated externally in which case the script wont run till the end.

Additional context
high level frameworks like keras and pytorch-lightning have csvlogger that attaches to a trainer. MONAI has MetricLogger that logs but does not save to a csv. MONAI MetricsSaver is may be what we need.