fix 06-inter-node-reduce-scatter.py tutorial bug

[AkiRusProd]

Hello!

In the create_reduce_scater_2d_ctx method of 06-inter-node-reduce-scatter.py tutorial. the global process rank was used to index the signal_bufs list during the initialization of signal_buf.

This resulted in an IndexError on any node other than the first (i.e., when rank >= local_world_size).

What is the solution?
The indexing for the signal_bufs list has been corrected to use local_rank instead of the global rank:

- signal_buf = signal_bufs[rank]
+ signal_buf = signal_bufs[local_rank]

This change ensures that each process on a node correctly accesses its dedicated signal buffer from the list of local buffers. It resolves the crash and enables the proper initialization of the communication context for multi-node environments.

NOTE:
from nvshmem_create_tensors:

rank_on_same_node_start = rank - local_rank
rank_on_same_node_end = rank_on_same_node_start + local_world_size

so then
len(signal_bufs) = rank_on_same_node_end - rank_on_same_node_start = (rank_on_same_node_start + local_world_size) - rank_on_same_node_start = local_world_size

[CLAassistant]

All committers have signed the CLA.