Skip dimension of size 1 in contiguous checks. (triton-inference-server#281)

* Skip dimension of size 1 in contiguous checks.

Since PyTorch 1.13 dimension of size 1 have normalised strides in PyTorch which fail here when using DLPack.
This was done to conform the torch stride representation and the one from numpy.
Unfortunately this means we are stuck with PyTorch 1.12.0 in python models.

* Conform to pre-commit guidelines

Author: MatthieuToulemont

src/pb_tensor.cc

@@ -433,12 +433,14 @@ PbTensor::FromDLPackCapsule(
     int64_t calculated_stride{1};
     bool is_contiguous_c_order = true;
     for (size_t i = 1; i < dims.size(); i++) {
-      if (strides[ndim - i] != calculated_stride) {
-        is_contiguous_c_order = false;
-        break;
-      }
+      if (dims[ndim - i] != 1) {
+        if (strides[ndim - i] != calculated_stride) {
+          is_contiguous_c_order = false;
+          break;
+        }
 
-      calculated_stride *= dims[ndim - i];
+        calculated_stride *= dims[ndim - i];
+      }
     }
 
     if (!is_contiguous_c_order) {