[Feature] Add distributed module (#59)

* [Feature] Add distributed module

* fix IS_DIST error

* all_reduce_dict does operations in-place

* support 'mean' operation

* provide local group process

* add tmpdir argument for collect_results

* add unit tests

* refactor unit tests

* simplify steps to create multiple processes

* minor fix

* describe the different of *gather* in mmengine and pytorch

* minor fix

* add unit tests for nccl

* test nccl backend in multiple gpu

* add get_default_group function to handle different torch versions

* minor fix

* [Feature] Add distributed module

* fix IS_DIST error

* all_reduce_dict does operations in-place

* support 'mean' operation

* provide local group process

* add tmpdir argument for collect_results

* add unit tests

* refactor unit tests

* simplify steps to create multiple processes

* minor fix

* describe the different of *gather* in mmengine and pytorch

* minor fix

* add unit tests for nccl

* test nccl backend in multiple gpu

* add get_default_group function to handle different torch versions

* minor fix

* minor fix

* handle torch1.5

* handle torch1.5

* minor fix

* fix typo

* refactor unit tests

* nccl does not support gather and gather_object

* fix gather

* fix collect_results_cpu

* fix collect_results and refactor unit tests

* fix collect_results unit tests

* handle torch.cat in torch1.5

* refine docstring

* refine docstring

* fix comments

* fix comments