Add checkpoint tagger utility (#1697)
* Add checkpoint tagger utility
* Add a checkpoint tagger utility class to keep track of checkpoints
* Can remove old/obsolete checkpoints
* Can be saved/loaded so enables easy restarts of training jobs
* Can interact with local disk and gcs
* Added utility functions to read/write from disk or gcs.
* Made the metrics_wrapper use those functions
* Address Davide s suggestions
* Added doc to gcsfs functions, added api to ChptTagger for getting tags, cleaned up function names
* Added api to get all tags from the tagger and applied s/outfile/fd/