Check if graph transposition impacts parsing

There are still some lingering issues with using original vs. transposed graphs, e.g.,
```
>>> predicates = ["rdfs:subClassOf"]
>>> tp = oi.termset_pairwise_similarity(["BFO:0000006"], ["BFO:0000018"], predicates)
Graph contains multiple disconnected components. Will ignore all but the largest component. 24 components are present. Largest component has 35 nodes.
>>> oi = get_implementation_from_shorthand("grape:sqlite:obo:phenio")
>>> tp = oi.termset_pairwise_similarity(["eye"], ["eyelash"], predicates)
Graph contains multiple disconnected components. Will ignore all but the largest component. 31982 components are present. Largest component has 242712 nodes.
Traceback (most recent call last):
File "", line 1, in
File "/home/harry/oakx-grape/src/oakx_grape/grape_implementation.py", line 324, in termset_pairwise_similarity
resnik_model = self._make_grape_resnik_model(dag=dag)
File "/home/harry/oakx-grape/src/oakx_grape/grape_implementation.py", line 287, in _make_grape_resnik_model
resnik_model.fit(graph, node_counts=counts)
File "/home/harry/oakx-grape/.venv/lib/python3.9/site-packages/embiggen/similarities/dag_resnik.py", line 32, in fit
self._model.fit(
ValueError: The current graph instance Unnamed is not directed acyclic.
```
This suggests that the graph isn't being interpreted as a DAG because the edges are going in the wrong direction - an issue that transposition will fix unless the original graph is already "correct".
Graphs should only be transposed if necessary, not by default.

Start in src/oakx_grape/grape_implementation.py, especially termset_pairwise_similarity and _make_grape_resnik_model, and reproduce the reported calls with the shown predicates and shorthand implementations. Trace whether the graph passed to embiggen's DAG Resnik model is directed correctly; done means transposition is conditional rather than applied by default and the reported non-DAG failure is addressed.

Written by the indexing model from the issue text.