Freeze calibrated verifier and prepare Unseen-v1 evaluation

[samarth1412]

What changed

• freezes the existing semantic verifier as semantic_v1_calibrated and adds versioned public artifact schemas and rule-pack boundaries;
• completes strict privacy controls, streaming-safe bounded FastAPI capture, and deterministic resource cleanup;
• adds bounded/batch verification and concurrent run isolation for LangChain, LangGraph, and LlamaIndex;
• prepares 1.1.0rc1 with Python 3.10--3.13, optional-integration, 80% coverage, dependency-audit, OIDC, attestation, and cross-platform wheel gates;
• defines the ContextTrace-Unseen-v1 acquisition, annotation, sealing, and preregistration protocol and updates overlapping related work.

Why

Existing RAGTruth, Diag-150, and Naturalistic Eval v2 results were repeatedly inspected during verifier development and must be treated as calibration evidence. This change freezes that implementation before collecting a source-family/domain/publication-window-disjoint test corpus.

Validation

• clean committed worktree: 483 tests passed;
• 85.90% coverage with an 80% CI floor;
• zero ResourceWarnings under warning-strict execution;
• Ruff correctness checks and scoped public-contract type checks passed;
• pip-audit reported no known vulnerabilities;
• 1.1.0rc1 wheel and sdist built and installed successfully in a clean environment;
• paper compiled successfully with the new bibliography.

Follow-up

Publish 1.1.0rc1 to TestPyPI, then acquire and freeze ContextTrace-Unseen-v1 before implementing or scoring semantic_core_v2.