Stage D: apply the two must-fix resolution-floor / resolved-tag constants

[WilfordGrimley]

Summary

• Owner-ruled must-fix (2026-07-22, docs/pipeline-fidelity-gate.md SS3): applies the two remaining MISSING knowledge-inventory constants to Stage D's _eligible_cards_queryset (local_calculate_verdicts.py), shared by both the join-key and fallback calculators.
• RESOLUTION_FLOOR_DPI = 200 excludes cards whose source image sits below the pilot's empirically-validated resolution floor. Implemented as .exclude(Q(dpi__lt=200) & Q(dpi__isnull=False)), not a bare .exclude(dpi__lt=200) — the bare form is not null-safe (NOT (dpi < 200) evaluates to NULL, not TRUE, for a NULL dpi under SQL's three-valued logic, so it would silently drop a null-dpi card too). Card.dpi is a DB-level NOT NULL column today (confirmed live: 0 nulls), so this only matters if that constraint ever moves — guarded anyway per the task spec.
• EXCLUDED_RESOLVED_TAGS = ["custom-art", "non-english"] excludes cards already tagged either, via the same tags__contains mechanism the live pilot's own _eligible_base_queryset uses.
• Constant Simplify flag icons: static SVG assets instead of hand-rolled geometry #3 (the deductive-backfill exclusion) is unaffected — it was separately ruled NOT restored the same day (PR Add Stage D fallback channel calculator (pre-fire prep) #341) and this PR does not touch that decision.
• Sizing (verified live 2026-07-22T23:47Z, cited in the task): the two excludes remove 28 (dpi) + 47 (custom-art) + 0 (non-english) = 75 non-overlapping cards from the current 179,766-card eligible pool (0.042%).
• Both constants are duplicated as module-level literals in local_calculate_verdicts.py (not imported from local_identify_printing_tags), matching this module's own established "avoid a hard import-time dependency between sibling engines over one constant" convention.
• See docs/pipeline-fidelity-gate.md SS3 for the full gate status this closes out (not edited in this PR — Dumont owns that doc this cycle in Ratify pipeline-fidelity gate artifact-1 parity-replay outcome #342).

Test plan

• pytest cardpicker/tests/test_local_calculate_verdicts.py — 84 passed (new TestEligibleCardsQueryset class: below-floor excluded, dpi==200 boundary included, null-dpi included, custom-art/non-english tagged excluded, untagged/differently-tagged included)
• pytest cardpicker/tests/test_local_calculate_verdicts.py cardpicker/tests/test_review_cluster_views.py cardpicker/tests/test_review_clusters.py cardpicker/tests/test_reparse_collector_evidence.py cardpicker/tests/test_local_detect_ai_art.py cardpicker/tests/test_local_identify_printing_tags.py — 413 passed (every module that touches this queryset or its sibling)
• pre-commit (ruff/isort/black/mypy/prettier) — all green on the commit
• Full-catalog Stage D fire — separate, owner-gated prod step, not part of this PR

🤖 Generated with Claude Code