PolicyEngineUKDataset.load() raises KeyError: 'person_weight' for directly-supplied UK datasets

Constructing PolicyEngineUKDataset(name=..., description=..., filepath=..., year=...) against any published UK microdata file raises KeyError: 'person_weight'.

Reproduction (with any policyengine-uk-data Enhanced FRS file):

import policyengine as pe
from policyengine.tax_benefit_models.uk.model import PolicyEngineUKDataset

ds = PolicyEngineUKDataset(name="x", description="x", filepath=PATH, year=2024)
# -> KeyError: 'person_weight'

Cause: load() in src/policyengine/tax_benefit_models/uk/datasets.py assumes person_weight and benunit_weight are already columns. Published UK microdata carries only household_weight at the household level (verified absent in enhanced_frs_2024_25.h5, enhanced_frs_2023_24.h5 and populace_uk_2023.h5); person and benunit weights are derived. create_datasets in the same file already derives them and saves the rewritten file, so the manifest path works while the direct-filepath path does not.

Fix: factor the derivation into a shared helper and call it from load() too, deriving only when the columns are absent.

Start in src/policyengine/tax_benefit_models/uk/datasets.py by reading load() and create_datasets(), then reproduce the direct-filepath failure with a published Enhanced FRS file. Trace the existing weight derivation and verify that both direct loading and the manifest path work, including files where the derived columns are already present.