[S2] Refactor pipeline (core/) has no per-component test suite

[erinepshovel-code]

The Backend production pipeline (backend/src/edcmbone/) has ~50 test methods covering canon, parser, stats, risk, compute, projection, matrix, compress, and closed_tokens.

The refactor pipeline (engine.py + core/) has:

• tests/test_smoke.py — one smoke test asserting only that the output is non-empty (structural shape, not value correctness).
• tests/run_all.py — a golden-drift harness that compares full JSON output against a checked-in baseline. The tests/golden/ directory appears empty, so the baseline may not yet exist; the first run would create it rather than verify against it.

No per-component tests exist for:

• core/parsing/normalizer.py
• core/parsing/tokenizer.py
• core/parsing/turn_builder.py
• core/parsing/round_builder.py
• core/operator/matcher.py
• core/operator/operator_extractor.py
• core/behavioral/behavioral_metrics.py
• core/behavioral/behavioral_window.py
• core/behavioral/structural_proxies.py
• core/bridge/bridge_engine.py
• compress.py (refactor-path decode)

Impact: Bugs in core/ are invisible until they corrupt end-to-end golden output, and then only if the golden baseline exists. All findings from the current code review (#47–#52) that touch core/ modules are currently untested in CI.

Fix: Add a tests/test_core_*.py suite mirroring the structure of Tests/test_backend.py. At minimum cover the happy path and one boundary condition per module.