Add extend_single_year_dataset for fast dataset year projection

[anth-volk]

Motivation

The API v2 alpha's create_datasets() is extremely slow (~1hr+ per state) because it routes every variable through sim.calculate(), invoking the full simulation engine for each variable × each year. The UK avoids this entirely: policyengine-uk has extend_single_year_dataset() which uprates DataFrames via simple multiplication — no simulation engine needed.

Now that policyengine-us-data is being updated to publish entity-level HDFStore files alongside the existing h5py files (PolicyEngine/policyengine-us-data#567), policyengine-us needs the machinery to consume them.

Changes

1. Dataset schema classes (data/dataset_schema.py)

• USSingleYearDataset — holds one DataFrame per entity (person, household, tax_unit, spm_unit, family, marital_unit) + time_period. Supports load/save/copy.
• USMultiYearDataset — holds a dict of USSingleYearDataset indexed by year. load() returns {variable: {year: array}} compatible with policyengine-core's TIME_PERIOD_ARRAYS format.

2. extend_single_year_dataset() (data/economic_assumptions.py)

Copies base-year DataFrames for each year through end_year (default 2035), then applies multiplicative uprating:

• Reads uprating parameter path from system.variables[var].uprating at runtime
• Computes growth factor as param(current_year) / param(prev_year) (absolute index ratio)
• Variables without uprating are carried forward unchanged
• No separate uprating list to maintain — picks up changes from variable definitions and default_uprating.py automatically

3. Dual-path loading in Microsimulation (system.py)

• _is_hdfstore_format() — detects entity-level HDFStore vs variable-centric h5py by checking top-level HDF5 keys
• _resolve_dataset_path() — resolves HuggingFace URLs and local paths
• Microsimulation.__init__ — when an HDFStore file is detected, loads it as USSingleYearDataset, extends via extend_single_year_dataset, and passes the resulting USMultiYearDataset to the parent class
• Legacy h5py files continue to work via the existing code path

Depends on

• Add entity-level HDFStore output format alongside h5py policyengine-us-data#567 (HDFStore output format)