Expand `Nmt_Paratext` E2E test to cover more SF use cases

[Enkidu93]

We've encountered a few problems recently that would have been caught had we been testing corpus set-ups more like SF (one parallel corpus for training another for inferencing). We could expand unit and integration tests to include this kind of set-up as well, but minimally, we should expand our E2E test to incorporate this arrangement. We should also exercise the USFM endpoint options in the E2E test to confirm further that quotation denormalization and marker placement are working as expected E2E. This is cheaper than semi-manual tests by the test team and will give us some confidence that last year's major draft improvement features are not regressing when used in a SF-like corpora set-up.