Fix Enhanced CPS PUF clone calibration guards

[MaxGhenis]

Summary

• Adds high-weight Enhanced CPS calibration targets for the source household total and a 50/50 CPS vs PUF-clone household split.
• Seeds CPS and PUF-clone household priors deterministically so each half starts with half the source household count before calibration.
• Excludes Forbes-marked PUF rows from QRF clone training; when metadata is missing, excludes all-record top-tail donors at $10M+ in AGI or selected financial components.
• Treats all-default Forbes metadata columns as missing metadata, so fallback PUFs do not silently use the Forbes-specific $250M filter.
• Keeps poverty as a post-build QA guard only; this PR does not add any poverty calibration target.
• Refreshes policyengine-us to 1.715.2 from upstream/main to resolve the merge conflict and satisfy freshness checks.

Local validation

• Full local 2024 Enhanced CPS rebuild completed under policyengine-us==1.715.2 at commit 192edf1e.
• Final guard output: 134.6M weighted households vs 135.3M source households, 49.9% PUF-clone household share, 19.3% person poverty QA, 31,670 nonzero weights.
• MicroSeries headline sums: LTCG $719.5B, household market income $16.50T, income tax $2.56T, SPM taxes $4.13T.
• Capital gains residuals: CBO loss-limited net capital gains $950.7B vs $1.291T (-26.4%); SOI long-term capital gains $719.5B vs $1.274T (-43.5%).
• Full target residual sweep: mean squared relative error 0.465, median 0.00278, 71.2% of 3,706 targets within 10%.

Checks

• PYTHONUNBUFFERED=1 uv run python -m policyengine_us_data.datasets.cps.enhanced_cps
• uv run ruff format --check .
• uv run ruff check policyengine_us_data/calibration/puf_impute.py policyengine_us_data/datasets/cps/enhanced_cps.py policyengine_us_data/utils/loss.py policyengine_us_data/utils/__init__.py tests/unit/calibration/test_calibration_puf_impute.py tests/unit/calibration/test_loss_targets.py tests/unit/datasets/test_enhanced_cps_seeding.py
• uv run pytest tests/unit/calibration/test_calibration_puf_impute.py tests/unit/calibration/test_loss_targets.py tests/unit/datasets/test_enhanced_cps_seeding.py -q (83 passed)
• git diff --check
• Full local target residual sweep script over build_loss_matrix(LocalEnhancedCPS, 2024)