Quarantine or fix known-broken tests on main (Blosc2, AskRequest)

从 tests/to_refactor/python/test_us_policy_macro.py::test_utah、tests/test_household_calculation.py、tests/test_agent.py 和 app/src/tests/unit/libs/migration/* 开始；运行受影响的测试套件，以确认列出的失败。检查每个 repository 现有的测试和 CI 配置，然后确保每个已知失败要么得到修复，要么被明确隔离并单独呈现，同时不掩盖回归问题。

Summary

Several test failures are known-broken on main across repos but are neither fixed nor quarantined. Every PR body ends up noting "these failures predate this PR" instead of the suite being green. This makes real regressions harder to spot.

What goes wrong

policyengine-api

• tests/to_refactor/python/test_us_policy_macro.py::test_utah hangs locally; module references policyengine_api/worker.py that no longer exists. Confirmed on master by multiple agents.

policyengine-api-v2 (simulation project)

• tests/test_household_calculation.py — macOS-only Blosc2 import failures.
• tests/test_agent.py — AskRequest ImportError.

policyengine-app-v2

• 40 failures in app/src/tests/unit/libs/migration/* from localStorage.clear is not a function.

Suggested fix

Per repo, either:

1. Fix the underlying issue (preferred), or
2. Quarantine with an explicit marker so the default suite is green:

# pytest
@pytest.mark.skip(reason="Broken on main — tracked in #NNNN, fix pending")
def test_utah(): ...

# vitest
it.skip('migration test — localStorage.clear not a function, tracked in #NNNN', ...)

3. Surface via a separate CI job (pytest -m broken) that's allowed to fail, so the status is visible without masking real regressions.

The anti-pattern to eliminate: PR authors noting "these failures predate this PR" in every review. If they predate every PR, they should be marked so CI reflects that.

Severity

Medium (dev-experience). Hides real regressions, creates review friction, breaks the "CI green = merge" invariant.

Scope

Filing this in three repos: policyengine-api, policyengine-api-v2, policyengine-app-v2.