Teardown in ClusterEnviroment

[four4fish]

Teardown behavior in Cluster Environment is not consistent. Is this by design?

1. Some environment has teardown logic to delete os.env flag, some doesn't. But in TTP, setup environment config logic applied to any cluster_enviroment

   In DDP:
   https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/plugins/training_type/ddp.py#L193-L202

   In ClusterEnviroment:
   https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/plugins/environments/cluster_environment.py#L72-L74

   In LightningEnviroment:
   https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/plugins/environments/lightning_environment.py#L80-L82

   All other environment don't have tear down logic override.

2. Only world size os env has been deleted, but not master port, master_address, rank, global_rank and etc.
   For example, several os.env has been set in DDP:
   https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/plugins/training_type/ddp.py#L197-L202

   In LightningEnviroment:
   https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/plugins/environments/lightning_environment.py#L80-L82