Describe the issue
Using the built-in Python bindings, certain quad and mixed-arity meshes produce zeros in place of correct values in numpy_num_face_vertices(), while the values in num_face_vertices are correct.

To Reproduce
Run the following Python code:

from tinyobjloader import ObjReader, ObjReaderConfig

reader = ObjReader()
config = ObjReaderConfig()
config.triangulate = False
reader.ParseFromString(mesh_string, "", config)
shapes = reader.GetShapes()
print(shapes[0].mesh.numpy_num_face_vertices())

Load either of the following meshes:

v 0 0 0
v 0 0 0
v 0 0 0
v 0 0 0
f 1 2 3 4
v 46.367584 82.676086 8.867414
v 46.524185 82.81955 8.825487
v 46.59864 83.086678 8.88121
v 46.461926 82.834091 8.953863
f 5 6 7 8

v 0 1 1
v 0 2 2
v 0 3 3
v 0 4 4
v 0 5 5
f 1 2 3 4
f 1 4 5

Expected behavior
The code above produces array([4, 0], dtype=uint8) for both meshes. For the first mesh the expected result is array([4, 4], dtype=uint8) and for the second mesh the expected result is array([4, 3], dtype=uint8).

With config.triangulate = True, the first mesh produces array([3, 0, 0, 0], dtype=uint8) whereas the expected result is array([3, 3, 3, 3], dtype=uint8). The second mesh the produces array([3, 0, 0], dtype=uint8) whereas the expected result is array([3, 3, 3], dtype=uint8).

In all cases the values of shapes[0].mesh.num_face_vertices are correct.

Environment

• TinyObjLoader 2.0.0rc12 and 2.0.0rc11
• OS: Linux and Mac
• Compiler Using official wheels published to pypi
• Other environment Python 3.9, 3.10; NumPy 1.21.1