fix(zero): warn when process group silently falls back to single-rank

[NovusEdge]

Summary

Adds a warning when zero.Init detects that WORLD_SIZE env var suggests multi-GPU but the actual process group has world_size=1.

This catches the sneaky case where torch.distributed wasnt initialized before model construction, so each rank loads the full model instead of sharding -> boom, OOM :D

The problem

When using deepspeed --num_gpus 8 but calling from_pretrained() before deepspeed.init_distributed(), zero.Init silently falls back to single-rank mode. No warning, just an OOM that looks like "model too big" when it actually fits fine if sharded properly.

The fix

Check if WORLD_SIZE > 1 but dp_world_size == 1 after process group setup, and warn loudly with a clear explanation of whats happening and how to fix it.

Closes #8084

[tohtana]

Hi @NovusEdge,

Thank you for submitting this PR!
However, I found #8089 seems to supersede this PR. As #8089 is submitted earlier, let me close this PR.

I also see that the predicate in the current approach env_world_size > 1 and self.dp_world_size == 1 is too broad.
self.dp_world_size is computed from self.ds_process_group, and that group can be an explicit data_parallel_group or sequence_data_parallel_group, not necessarily the world group. A multi-rank job can intentionally use a singleton data-parallel group when another parallel dimension is enabled.
I think #8089 can handle this properly.