Expose ability to create barriers in OpTensor operations

[axsaucedo]

Currently the optensor operations explicitly do not create a barrier, but there may be situations where this may be desired. Identify whether there are usecases where this would be relevant and explore best ways to expose the barrier creation (whether as a constructor parameter, or potentially as a completely separate default operation).

[axsaucedo]

It will be important to ensure the barriers are created, or that at least explicit requirements are made intuitive or explicit to the users in situations where it would cause issues, such as the OpTensor commands such as sync or copy, where the record step woudl perform a copy, and the postSubmit would perform a mapdatafromhost. This would work as long as there is a cpu barrier in the sequence submit, but if there isn't this would cause issues (it would involve fence and barriers).