OBJ reader fix

[andrsd]

Closes #953

[andrsd]

@quocdang1998 This fixes the bug in the OBJ reader. However:

1. parmetis does not seem to do its job and then we fail inside PETSc with adjacency issues. This is why I have the KBA partitioner in the example/test.
2. I see:

   *** WARNING ***  Cell quality checks detected 210 negative volume sub-elements (sub-triangle or sub-tetrahedron). This issue could result in incorrect quantities under some circumstances.
   

   This may indicate that the orientations of the elements are not correct. So, you may want to double check that.

Also pulling in @wdhawkins and @ragusa

[andrsd]

If you guys are ok with this, there is small refactoring I want do in the reader. Some of the parsing code would deserve a bit of improvement...

[andrsd]

Made 2 changes:

1. the way the parser works: we read a line, split it and process the parts rather than splitting with the same code inside the "subblock"
2. when conversion from string to a number fails, we stop now. The old version reported a warning and continued. the user would see:

   [0]  Making Unpartitioned mesh from wavefront file ../../../assets/mesh/cube.obj
   [0]  *** WARNING ***  Failed to convert vertex in line v a1.000000 -1.000000 -1.000000
   [0]  Max material id: 0
   [0]  **** ERROR ****  RuntimeError: Cannot normalize a zero-length vector.
   [0]  **** ERROR ****  
   [0]  **** ERROR ****  At:
   [0]  **** ERROR ****    read_wavefront_obj2.py(20): <module>
   

   Now they see:

   [0]  Making Unpartitioned mesh from wavefront file ../../../assets/mesh/cube.obj
   [0]  **** ERROR ****  RuntimeError: Failed to convert string to floating-point number in ../../../assets/mesh/cube.obj, line 5
   [0]  **** ERROR ****  
   [0]  **** ERROR ****  
   [0]  **** ERROR ****  At:
   [0]  **** ERROR ****    read_wavefront_obj2.py(20): <module>