Retain Swift Data summary precision blocker

[DavidBakerEffendi]

The independent Swift Data/Base64/String control finds a false flow through constant-body local lookalikes. Retain the blocked result: the canonical summary pair remains unrun and no scored configuration is activated.

Key Changes:

• Preregister and retain real raw-input, real safe-literal, and constant-body local roundtrips. Keep resolved declaration modules/owners distinct from import spelling and from the nominal owner of local String extensions.
• Retain attached manual ports, simple call-site edges, and a separately preregistered structural store witness. The store probe covers OptionalSome, which the simple-edge primitive explicitly excludes; empty origins do not claim exact catalog-row attribution. Add portable integrity and mutation checks preserving the false flow and blocked qualification.

Touch Points:

• adapters/codeql/swift-summary-v1/queries/
• evidence/swift-summary-qualification-v1/
• scripts/run-swift-summary-control.py, scripts/trace-swift-summary-stores.py
• scripts/verify-swift-summary-control.py, scripts/test-swift-summary-control.py
• .github/workflows/ci.yml, .gitattributes, Swift qualification documentation

Validation: three test methods with seven semantic and four copied-evidence mutations passed, including from a clean Git export. Previous sanitizer/append/pending-integration verifiers, cargo fmt --check, and diff checks passed. Extraction completed in 111.512 seconds; all six original queries and the 13.034-second store query succeeded under the 150-second extraction / 60-second query / 2048 MiB diagnostic request. The reused database's 422 non-cache source-data hashes and source archive were checked. Aggregate memory compliance and semantic completeness remain unproven. CI pending.

Refs #220 and #215. Both issues remain open; acceptance is incomplete. Append and sanitizer blockers remain independent. No full-corpus execution, scored activation, freeze, or publication.