Saved UK outputs are reused after their year file is re-cut #606

Problem

A saved UK simulation output is checked only for a recorded data year. MicrosimulationModelVersion.load() (src/policyengine/tax_benefit_models/common/model_version.py) calls uk.datasets._year_file_records_data_year and nothing else.

#601 (issue #600) stamps every UK year file with the source dataset it was cut from, the policyengine-core and policyengine-uk versions that cut it, its year and a format revision (uk.datasets.YEAR_FILE_STAMP_H5_DATASET), and ensure_datasets re-cuts a year file whose stamp is out of date. Nothing ties a saved output to that stamp. So when a year file is re-cut, because the bundle re-pinned the dataset, the local source was edited or policyengine-uk was upgraded, Simulation.ensure() and Simulation.load() still reuse outputs calculated from the old year file.

#594 (issue #592) closes the same gap for US outputs: a run records an output stamp holding the input year file's stamp, save() writes it after the tables, and load() refuses an output whose stamp is out of date.

Fix

Mirror #594 for the UK:

• A UK dataset read from a year file, or returned by create_datasets, keeps the file's stamp.
• A UK Simulation.run() records an output stamp holding that stamp and an output format revision.
• save() writes the output stamp after the tables, in its own H5 dataset.
• Simulation.load() refuses, and Simulation.ensure() recalculates, an output whose stamp is not this release's, or which was calculated from a year file other than the simulation's dataset holds now, or from one cut under another year-file revision or other package versions.

Stacked on #601.