fix: preserve hierarchical indexing estimate rules

[astordu]

Summary

• Preserve parent-child indexing estimate fields during request validation
• Keep parent_mode and subchunk_segmentation when normalizing process rules
• Add a regression test for hierarchical indexing estimate validation

Problem

When creating a knowledge base with parent-child chunking, the preview endpoint can return zero chunks even though the uploaded document is parsed correctly. The request payload includes parent_mode and subchunk_segmentation, but DocumentService.estimate_args_validate rebuilds process_rule from _EstimateRules, which previously only allowed pre_processing_rules and segmentation. As a result, parent-child specific fields are dropped before indexing estimation.

Test

• python -m pytest -o addopts='' tests/unit_tests/services/test_dataset_service_document.py -k estimate_args_validate_preserves_hierarchical_chunking_rules -q