添加MPI并行TN benchmark及辅助脚本，移除旧benchmark

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

inspect_path.py

@@ -0,0 +1,21 @@
+import pickle
+import sys
+
+path = sys.argv[1] if len(sys.argv) > 1 else 'path/path.pkl.34.mpi'
+
+with open(path, 'rb') as f:
+    d = pickle.load(f)
+tree = d['tree']
+
+width = tree.contraction_width()
+slices = tree.multiplicity
+sliced_width = width - (slices.bit_length() - 1)
+
+print(f"Path: {path}")
+print(f"Width (unsliced):     {width:.1f}")
+print(f"Slices:               {slices}")
+print(f"Sliced width:         {sliced_width:.1f}")
+print(f"Peak memory (total):  {2**width * 16 / 1e9:.1f} GB")
+print(f"Peak per slice:       {2**sliced_width * 16 / 1e9:.2f} GB")
+print(f"Sliced indices:       {len(tree.sliced_inds)}")
+print(f"Cost (log2 flops):    {tree.contraction_cost(log=True):.2f}")