Stop defaulting Simulation() to a private dataset

[MaxGhenis]

Summary

• remove the hardcoded private Hugging Face dataset from bare Simulation() / Microsimulation() initialization
• require an explicit dataset when no situation is provided, unless POLICYENGINE_UK_DEFAULT_DATASET is explicitly set
• add regression tests covering both the explicit error path and the opt-in env-var path

This is a corrective follow-up to the merged default-dataset change. It avoids silently reaching into a private HF repo from the public package.

Verification

• pytest policyengine_uk/tests/test_deterministic_variables.py::TestDefaultDatasetUrl -q