feat(preflight): evaluate optional-content membership for print and screen (#115)

[mberrys]

Closes #115

What changed

hidden-layers now evaluates every optional-content governor it meets under both the View and the Print usage (default configuration with its /AS events applied) and reports what it finds:

• Governors covered: groups and membership dictionaries (/P AnyOn, AllOn, AnyOff, AllOff and /VE expressions) on marked content (BDC, including inline dictionaries the content processor cannot evaluate), Form XObject /OC entries (Do), and annotation /OC entries.
• Reported: membership hidden in print, and content whose screen and print visibility differ, in either direction, with evidence.view_state, print_state, policy, ocg_names, governor, governor_source, divergence.
• A governor that cannot be evaluated (unparseable expression, unknown group state) is an error finding with inspection_complete: false, not a pass.
• Overlay row hidden-layers-ocmd is landed (closed by hidden-layers); hidden-layers measures, limitations and evidence rewritten.

Existing behaviour is kept: a plain group hidden in print from marked content is still reported once by name, and the golden corpus is unchanged.

Proof

Source SHA under test: 2478cd80. Fixture: loop-preflight/testdata/fixtures/ocmd-print-divergence.pdf (generator loop-preflight/tools/generate_ocmd_divergence_fixture.py), committed first (38d7af0). It has two groups with opposite View/Print usage and five membership dictionaries (AllOn and AllOff on Form XObjects; AnyOn, AnyOff and /VE [/Not ...] on marked content), each diverging between screen and print.

	Before	After
ocmd-print-divergence (test-hidden-layers, warning)	pass: true, no errors, no warnings	5 hidden-layers warnings, one per policy; AllOn is on-screen/off-print, the other four off-screen/on-print
golden corpus (156 tests)		156/156 pass; the only snapshot change is the new ocmd-print-divergence.json

The baseline binary was built from the PR #156 head, which does not touch hidden-content code.

Local (Windows, MSVC Release): UnitTestsPreflightEngine 112/112 (includes the divergence test and an unevaluable-expression test that expects an incomplete error), UnitTestsPreflightCorpus 156/156, UnitTestsPreflightVerdict 53/53, UnitTestsPreflightChecks 17/17, UnitTestsEvidenceGraph 15/15, UnitTestsContentProcessorLimits 10/10. python scripts/generate-architecture-catalogs.py --check, check_source_integrity.py, check_preflight_truth_source.py, generate-adapters.py and check-change.py --dry-run pass.

Skipped or unavailable locally: the full check-change.py build/ctest/clang-tidy run (no clang-tidy-18; only the targets above were built), Linux lanes, and the remaining mapped test targets. CI covers those.

Remaining risk

• Alternate optional-content configurations are not evaluated, only the default.
• An /OC entry on a Form XObject or annotation that is never painted is not inspected.
• Annotations are inspected from the page's /Annots only when the annotation dictionary itself carries /OC.

Anti-slop review

No second visibility engine: evaluation reuses PDFOptionalContentMembershipObject and PDFOptionalContentActivity for both usages, and the only new decision logic is when to report. A governor is never silently treated as visible when it cannot be evaluated.

🤖 Generated with Claude Code

[mberrys]

Duplicate finding for marked-content OCMDs that are hidden in print. performMarkedContentBegin already adds a hidden-layers finding ("unnamed optional-content group") when isContentSuppressed() is true. This skip test only exempts plain OCGs, so the same OCMD is reported a second time here. A plain OCG in BDC that diverges between view and print is also reported twice. The fixture's BDC OCMDs are all print-ON, so the tests don't catch this. The marked-content path should defer to the governor report when the governor is an OCMD, or vice versa.