There is no such functionality in the SDK. Implementing it yourself would most likely mean keeping a list of twins/relationships added successfully and removing them if one addition fails. If you start with removing the added relationships first, you should have no trouble reverting to the initial state. But when you start reverting thousands of twins/relationships, this could become quite time-consuming.
One new feature to look into would be data importing from the CLI. There is a new feature where you can create a "job". That might support transactions, but the documentation is extremely limited, and I haven't tried it yet. The idea is that you upload a file to blob storage, containing models, twins and relationships. You give your ADT instance read/write rights on that container and start a job through the CLI. The job will then place an output file next to the uploaded input file with the results.