fix: support missing torch with spark and multiprocessing backends

[percevalw]

Description

Fixes #243

• one of the errors came from torch being imported without handling import errors (torch is an optional dependency)
• the second main error came from the from .registry import registry in edsnlp/core.py, preventing python from programmatically importing the classes from edsnlp/core/registry.py. Because of this, dill tryed to serialize the entire registry module, and some other modules from edsnlp, causing various serialization issues.

Checklist

• [x] If this PR is a bug fix, the bug is documented in the test suite.
• [x] Changes were documented in the changelog (pending section).
• [x] If necessary, changes were made to the documentation (eg new pipeline).

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Comparison is base (eb08237) 96.70% compared to head (2da3cc3) 96.70%.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #244 +/- ## ======================================= Coverage 96.70% 96.70% ======================================= Files 255 255 Lines 8369 8372 +3 ======================================= + Hits 8093 8096 +3 Misses 276 276 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.