add runnable data parallel example

[tylergibbs1]

proposed changes

#2930 asks for standalone distributed examples. #2973 added the explanation, but examples/python still has no data parallel script users can run end to end.

this adds a linear regression example where each rank creates its own synthetic data shard, starts from the same parameters, and averages gradients with nn.average_gradients. the final loss is evaluated on every rank before rank zero prints. without that step, lazy evaluation can leave one process waiting on a collective after its peer exits.

run it with:

mlx.launch -n 2 python examples/python/data_parallel.py

on my m4 mac, the one rank run finished with an l2 distance of 0.00971. the two rank run finished at 0.00464.

checks

• pre-commit run --all-files
• python examples/python/data_parallel.py
• mlx.launch -n 2 python examples/python/data_parallel.py

checklist

• i have read the contributing document
• i have run pre-commit run --all-files
• i have added automated tests. this pr adds a runnable example and the checks above exercise both paths
• i have updated the necessary documentation. no api or existing docs changed