Always use spawn for data loader workers

[jder]

Summary

• Remove fork/spawn selection based on data source location and always use spawn when data loader workers are enabled.
• Simplify DataContainer and location resolution by dropping fork-support checks that are no longer needed.
• Update trainer tests to assert the multiprocessing context is spawn for worker-backed loaders and None when workers are disabled.

Testing

• Added/updated unit coverage for trainer loader setup.
• Not run (not requested).