[BUG] String sort based on ``rank_repr`` may lead to problems?

Describe the bug
This is more of question than a known bug, and if there is a bug, it would probably only occur for very large models or extreme cases. There may be an issue based on how checkpoint file names are sorted.

The file names include a rank_repr encoding:

https://github.com/microsoft/DeepSpeed/blob/4ad8019cdf5b48ff169bacc2ca3dce5f80671709/deepspeed/runtime/pipe/module.py#L537

which uses two-digit formatting as shown here:

https://github.com/microsoft/DeepSpeed/blob/4ad8019cdf5b48ff169bacc2ca3dce5f80671709/deepspeed/runtime/pipe/topology.py#L99

So one gets a set of file names like:

layer_01-model_00-model_states.pt
layer_01-model_01-model_states.pt
layer_01-model_02-model_states.pt
layer_01-model_03-model_states.pt

where the model_00 part is generated by the rank_repr encoding.

When reading those checkpoint files back, they are returned in an ordered list, where a normal string sort is used to order the files, like here:

https://github.com/microsoft/DeepSpeed/blob/4ad8019cdf5b48ff169bacc2ca3dce5f80671709/deepspeed/runtime/pipe/module.py#L549

I think that could lead to problems if one were to use really large models, e.g., if the tensor parallelism reached 100+, since files would be ordered like so (where 100 comes before 11):

layer_01-model_09-model_states.pt
layer_01-model_10-model_states.pt
layer_01-model_100-model_states.pt
layer_01-model_11-model_states.pt

I believe that would mix weights, perhaps in a silent way that is hard for the user to detect. For example, this line might cause rank 11 to read in the weights from rank 100:

https://github.com/microsoft/DeepSpeed/blob/4ad8019cdf5b48ff169bacc2ca3dce5f80671709/deepspeed/runtime/state_dict_factory.py#L80

In the short term, perhaps an assert could be added to at least alert the user there is a problem if they exceed two digits on any rank_repr encoding?

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior
A clear and concise description of what you expected to happen.

ds_report output
Please run ds_report to give us details about your setup.

Screenshots
If applicable, add screenshots to help explain your problem.

System info (please complete the following information):

• OS: [e.g. Ubuntu 18.04]
• GPU count and types [e.g. two machines with x8 A100s each]
• Interconnects (if applicable) [e.g., two machines connected with 100 Gbps IB]
• Python version
• Any other relevant info about your setup

Launcher context
Are you launching your experiment with the deepspeed launcher, MPI, or something else?

Docker context
Are you using a specific docker image that you can share?

Additional context
Add any other context about the problem here.

Start with rank_repr in deepspeed/runtime/pipe/module.py and its formatting in deepspeed/runtime/pipe/topology.py, then trace checkpoint ordering in module.py and deepspeed/runtime/state_dict_factory.py. Reproduce the load path with rank values above 99 and verify that sorted checkpoint files map each rank to the correct weights; done means the issue is detected or ordering is shown to be safe.