Changing Reader.get_next_str (which returns a Python String)
into a c function saves a lot of time.
Specifically it avoids a GetAttr call, and a PyObject_CallObject
This drops the times down to:
...test__read_dirblocks_20k_tree_0_parents_c    OK      122ms/    2561ms
...test__read_dirblocks_20k_tree_0_parents_py   OK      235ms/    2606ms
...test__read_dirblocks_20k_tree_1_parent_c     OK      175ms/    2797ms
...test__read_dirblocks_20k_tree_1_parent_py    OK      358ms/    3014ms
...test__read_dirblocks_20k_tree_2_parents_c    OK      259ms/    2992ms
...test__read_dirblocks_20k_tree_2_parents_py   OK      498ms/    3232ms

We are close to being 2x faster than the python implementation.