Preserve flexible ELN metadata named parts

[CAOShurong]

Summary

• preserve flexible ELN metadata whose sanitized key is already parts
• choose a collision-free key for imported nested Dataset relationships using the existing ELN key sanitizer
• add an end-to-end regression based on the existing SciLog interoperability archive

Problem

Nested Dataset import currently adds generated object references under a hard-coded parts property after the object has been created. If variableMeasured already contains a PropertyValue whose sanitized key is parts, this post-processing step overwrites both its schema and data.

The importer already has _sanitize_keys() for resolving flexible-metadata key collisions. This change uses the same sanitizer for the generated relationship property. Existing imports without a collision keep the parts key; colliding imports preserve the original metadata and use property_parts for the generated references.

Verification

• Before the change, the new SciLog-based regression fails because Original parts metadata is replaced by five object references.
• python -m pytest -q --tb=short tests/logic/test_eln_import.py::test_import_nested_datasets_preserves_parts_metadata tests/logic/test_eln_import.py::test_import_nested_datasets
• Representative SampleDB, SciLog, eLabFTW, PASTA, and Kadi4Mat importer paths: 6 passed.
• Full Windows importer comparison: this branch 35 passed / 22 failed; upstream develop at d9374c2f 34 passed / 22 failed, with the same 22 platform/environment-dependent failures.
• Project CI static-analysis commands passed: pycodestyle, pyflakes, pylint, and strict mypy.

[FlorianRhiem]

Hey @CAOShurong, thanks for the PR, I've merged it into develop.