Sync random seed across ranks in distributed chat

[kernelpool]

This fixes an issue affecting distributed chat where a temperature > 0 would result in ranks having a different random state, causing them to sample different tokens and diverge (often seen as artifacts in the output, shown below).

This fix follows the approach used in #741 and 1d76aab

mlx.launch --verbose --backend jaccl --hostfile hosts-jaccl.json --env MLX_METAL_FAST_SYNCH=1 -- /Users/optimus/repo/mlx-lm/.venv/bin/mlx_lm.chat --model mlx-community/Qwen3-4B-Instruct-2507-8bit --temp 1.0
[INFO] Running /Users/optimus/repo/mlx-lm/.venv/bin/python /Users/optimus/repo/mlx-lm/.venv/bin/mlx_lm.chat --model mlx-community/Qwen3-4B-Instruct-2507-8bit --temp 1.0 
Fetching 10 files: 100% 10/10 [00:00<00:00, 106184.91it/s]
Download complete: : 0.00B [00:00, ?B/s]              
Fetching 11 files: 100% 11/11 [00:00<00:00, 38067.12it/s]
Download complete: : 0.00B [00:00, ?B/s]              
Fetching 10 files: 100% 10/10 [00:02<00:00,  4.07it/s]MB/s]                
Fetching 11 files: 100% 11/11 [00:49<00:00,  4.46s/it]     
Download complete: 100% 4.27G/4.27G [00:49<00:00, 86.9MB/s]00:00, 152MB/s]  
Download complete: : 15.9MB [00:52, 306kB/s] :00, 152MB/s]                
[INFO] Starting chat session with mlx-community/Qwen3-4B-Instruct-2507-8bit.
The command list:
- 'q' to exit
- 'r' to reset the chat
- 'h' to display these commands
>> hello!
Hello! � How can I assist you today?

[awni]

Good catch! But this ignores the seed flag above. Wdyt about instead just setting a default seed and using that?

[kernelpool]

Ah, makes sense

[awni]

Nit: you can just set the DEFAULT_SEED=0 so the behavior is consistent in all cases. I think it's fine if the default behavior is seeded.

[awni]

Thx!