Add test for distributed run

[ways]

Test fails. I think it's due to being on CPU(?). If so, the test should be marked clearly. Best solution would be to fix running on CPU.

[rank0]:   File "/home/larsfp/src/bris-inference/.tox/inference_CI_distributed/lib/python3.12/site-packages/anemoi/models/distributed/transformer.py", line 141, in forward
[rank0]:     return _headsalltoall(input_, shapes_, group=mgroup_)
[rank0]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/larsfp/src/bris-inference/.tox/inference_CI_distributed/lib/python3.12/site-packages/anemoi/models/distributed/transformer.py", line 53, in _headsalltoall
[rank0]:     dist.all_to_all(output_list, input_list, group=group)
[rank0]:   File "/home/larsfp/src/bris-inference/.tox/inference_CI_distributed/lib/python3.12/site-packages/torch/distributed/c10d_logger.py", line 83, in wrapper
[rank0]:     return func(*args, **kwargs)
[rank0]:            ^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/larsfp/src/bris-inference/.tox/inference_CI_distributed/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py", line 4114, in all_to_all
[rank0]:     work = group.alltoall(output_tensor_list, input_tensor_list, opts)
[rank0]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: RuntimeError: Backend gloo does not support alltoall

[ways]

https://docs.pytorch.org/docs/main/distributed.html#backends
Seems to be supported in torch 2.8, not earlier.

2.8 is not available in pip, dockerhub. Seems to still be under development.

[ways]

https://docs.pytorch.org/docs/main/distributed.html#backends Seems to be supported in torch 2.8, not earlier.

2.8 is not available in pip, dockerhub. Seems to still be under development.

See #139 (comment)