Require PolicyEngine 6.3.4 for regional simulations

[anth-volk]

Fixes #3864

Blocking dependency

Do not merge or deploy until all of these are done:

• PolicyEngine/policyengine.py#599 is merged and policyengine 6.3.4 is published.
• uv.lock is regenerated against it.
• PolicyEngine/policyengine-sim-api#726 is merged and successfully deployed.

The runtime and Docker image already require policyengine[models]==6.3.4. Until it is published, uv lock --check and the image build fail by design.

Why 6.3.4: with 6.2.7, every SPM calculation on the certified ACS-local source that states and congressional districts use stops with SPM_COMPOSITION_REQUIRED. The source predates SPM measurement-scope and independence-role declarations. 6.3.4 adds backward compatibility for such legacy sources:

• ACS group quarters are outside the SPM measurement, following Census's ACS household-population method.
• The source's ASEC/PUF records take the certified SPM independence role from the national release.

Summary

• Require policyengine[models]==6.3.4 in the API runtime and the generic container image. Use the bundle's exact Core 3.33.0, US 2.2.1, UK 2.126.12 and SPM calculator 1.0.0; do not select country packages independently.
• Verify that the national US default remains populace_us_2024, and that states and congressional districts use the certified populace_us_2024_acs_local dataset through standard region_datasets metadata, not dataset_overlays.
• Update the exact installed-catalog expectations for 6.3.4's models.
• Support England's multi-value region filter in catalog extraction:
  • Flat list or tuple membership values are stored as compact JSON arrays in the existing text column.
  • Scalar values keep their previous representation.
  • Invalid, empty, nested or oversized membership values fail before database access.
• Assert that the real installed England filter keeps all nine English regions.
• Null poverty rates pass through. The simulation API (Clear out duplicate policies #726) now reports null for a poverty group with nobody inside the poverty measurement (PovertyRateValue). That is not a measured 0%. A new test confirms economic impact results keep those nulls when returned and stored. API v1 does not parse poverty fields, so no other code change is needed.
• Update the bundle changelog fragment.

Catalog compatibility

With 6.3.4's models (US 2.2.1, UK 2.126.12), the installed catalog contains:

Entity	Count
Variables	7,294
Parameter nodes	29,300
Parameters	104,158
Parameter values	1,194,256
Datasets	3
Regions	826

These were measured with the published 6.3.1 bundle, which installs the same model versions; 6.3.4 changes no model package. Exact count assertions remain, so future country-package changes require explicit review.

Validation

In an environment with the published 6.3.1 bundle (same Core, US, UK and SPM calculator versions as 6.3.4):

• The complete API unit suite passed (2,297 tests), including the new null-poverty test.
• The installed-catalog compatibility test, constants and catalog-extraction tests passed (70 tests).

On the branch:

• Changed-file Ruff format and lint, migration quality guards, and migration-contract export passed. The regenerated artifacts have no tracked changes.

Not yet run: the lockfile refresh and frozen install against 6.3.4, the database-backed integration suite, container builds and live deployment checks. All of these wait for publication.

Deployment scope

• Existing Flask/API v1 routes and compute delegation remain in place.
• The only response change is upstream: poverty rates from the simulation API may now be null for a group with nobody measured. That does not happen for whole states or districts in current data.
• App follow-up (not yet tracked): policyengine-app-v2 types these rates as numbers, and its percentage-change text would show NaN for such a group.
• No new environment variables or secrets. The image installs 6.3.4, whose only new runtime download is a certified file from the same public Hugging Face repository as the ACS-local source; anonymous access was verified. UK data access is unchanged.
• No route cutover, Alembic revision, database provisioning, service account or deployment target. No running infrastructure or traffic was modified.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 60.00000% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 45.25%. Comparing base (21ddf34) to head (3c7c0e7).

Files with missing lines	Patch %	Lines
policyengine_api/data/v2/catalog/extraction.py	60.00%	4 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3868      +/-   ##
==========================================
+ Coverage   45.22%   45.25%   +0.02%     
==========================================
  Files         183      183              
  Lines       11073    11085      +12     
  Branches     1905     1907       +2     
==========================================
+ Hits         5008     5016       +8     
- Misses       5644     5647       +3     
- Partials      421      422       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.