[codex] require public numeric bounds

[vvv214]

Summary

This draft PR removes exact data-derived min/max computation from the pipeline numerical derivation path.

• require public NumericalAttribute bounds before DP quantile derivation
• pass descriptor-provided public bounds from dataset_encoding.encode_dataset
• fail fast when a numerical attribute lacks public bounds
• update tests and the directly related docs so the pipeline no longer claims to privately infer numerical bounds

Why

numerical_values_derivation.derive_numerical_attributes previously used pipeline_functions.min_max_per_key to derive exact min/max from sensitive data, then reused those exact values to normalize and denormalize DP quantiles. That leaks the extrema outside the DP mechanism.

Validation

• python3 -m py_compile dpsynth/pipeline_transformations/numerical_values_derivation.py dpsynth/pipeline_transformations/dataset_encoding.py tests/pipeline_transformations/numerical_values_derivation_test.py tests/pipeline_transformations/dataset_encoding_test.py
• rg -n "min_max_per_key|pipeline_functions|min_value=min_max|max_value=min_max|privately deduce domain boundaries|real-world data bounds" dpsynth/pipeline_transformations docs tests/pipeline_transformations/numerical_values_derivation_test.py tests/pipeline_transformations/dataset_encoding_test.py returns no matches

Full pytest was not run locally because this checkout's test dependencies (pipeline_dp, mbi, dp_accounting, Beam/JAX stack) are not installed in the current environment, and the local Python is 3.14 while the project declares <3.14.