Add pluggable region scoping strategies (RowFilter + WeightReplacement)

[anth-volk]

Problem

The refactored policyengine.py replaced the old region: str parameter with filter_field/filter_value on the Simulation model but only implemented row filtering. The weight replacement logic used for UK parliamentary constituencies and local authorities was lost. UK constituencies are defined with filter_field="household_weight" which doesn't work with the row-filter implementation.

Proposed solution

Introduce a RegionScopingStrategy abstraction with two concrete implementations:

• RowFilterStrategy: Filters dataset rows by matching a household-level variable (used for UK countries, US places)
• WeightReplacementStrategy: Downloads pre-computed weight matrices from GCS and replaces household_weight in the dataset (used for UK constituencies and local authorities)

Changes

• New scoping_strategy.py module with RegionScopingStrategy base class, RowFilterStrategy, WeightReplacementStrategy, and ScopingStrategy discriminated union type
• Updated Region and Simulation models with scoping_strategy field
• Updated UK and US model run() methods to use strategy pattern
• Updated UK region registry (constituencies + local authorities use WeightReplacementStrategy, countries use RowFilterStrategy)
• Updated US region registry (places use RowFilterStrategy)
• Comprehensive tests for strategy serialization, application, and model integration