[CuTeDSL] Distributed example, using TMA load to access remote memory rank-by-rank, reducing in cta, broadcast result to all ranks by multimem TMA store

[aragorn-guan]

Add TMA-based distributed all-reduce example (all_reduce_tma.py)

A tutorial example demonstrating TMA usage for distributed all-reduce operations across multiple GPUs.

Key features:

• Uses 1D TMA load to load from remote GPU memory via NVSHMEM addresses
• Uses 1D TMA store to store to multicast address for broadcasting
• Supports any input shape by flattening to 1D and tiling linearly
• Two-stage pipeline overlaps TMA loads across ranks

Note: This example prioritizes clarity over performance optimization, serving as a learning resource for TMA-based distributed operations.

[shubaoyu2]

LGTM，and also cc @IonThruster @brandon-yujie-sun @fengxie @hwu36 for review and approve