Potential issue with loading injection data for different backends

The following discussion from !36 (merged) should be addressed:

• @amanda.farah started a discussion: (+1 comment)

  Just a note that the new "backends" infrastructure seems to cause problems in vt_helper.py, at least when I pulled these changes two days ago. The problem comes in around line 171 (in load_injection_data.py)

      if vt_file.endswith(".pkl"):
          with open(vt_file, "rb") as ff:
              data = dill.load(ff)
          for key, value in data.items():
              if isinstance(value, np.ndarray):
                  data[key] = xp.asarray(value)
          return data

  I think this block of code is meant to change numpy arrays to cupy or jax arrays, but they always stay numpy arrays, which causes problems almost immediately downstream with cupy. If I print out xp and np, I always get <module 'numpy' from '/home/amanda.farah/.conda/envs/analyst_guidelines_2/lib/python3.10/site-packages/numpy/__init__.py'> regardless of what I set the backend argument to in my config file. I tried to fix this with a from gwpopulation.backend import set_backend and a set_backend(args.backend) in the relevant function, but this had no effect on xp or np. This seemed like it should have worked based on what set_backend() does, but I have to admit I don't fully understand how importlib.metadata.entry_points works.