BUG: fix scipy backend pickling after multiple opens

[hansu650]

Closes #11323.

This fixes a scipy backend pickling regression when multiple datasets are opened
from file-like objects in the same process.

The scipy backend registers a pickle-visible wrapper class for file-like objects.
Re-registering that class on every open can make an earlier dataset refer to a
class object that no longer matches the registered pickle target. This change
registers the wrapper once and reuses it for later file-like opens.

The regression test opens two scipy-backed datasets from the same in-memory
NetCDF payload and verifies that the first dataset can still be pickled and
unpickled.

Validation:

• python -m pytest xarray/tests/test_backends.py -k "scipy and pickle"
• python -m pytest xarray/tests/test_backends.py -k "scipy"
• python -m ruff check xarray/backends/scipy_.py xarray/tests/test_backends.py
• git diff --check

AI assistance:
I used ChatGPT/Codex to help inspect the issue and plan a small local patch. I reviewed the final diff and test results myself and understand the submitted changes.

[hansu650]

Thanks for the CI feedback. I'm going to close this for now because I don't have time to follow up on the Mypy failure properly. Sorry for the noise.