[S3] Dead-code and incomplete-plumbing sweep

[erinepshovel-code]

Six small cleanup items suitable for a single PR:

1. core/operator/matcher.py::hyphen_compound_emission — imported by operator_extractor.py but never called; hyphen handling is inlined at the call site. Dead import + dead function.

2. backend/src/edcmbone/parser/turns_rounds.py — import math present but math is never referenced in the file. Dead import.

3. core/parsing/tokenizer.py — dead fallback if not toks: toks = text.split() is unreachable: the trailing-gap path always appends at least one token. Dead branch.

4. canon/loader.py — affix section list is hardcoded: ["inflectional", "derivational_prefixes", "derivational_suffixes"]. A new section added to the JSON is silently ignored. Should iterate affix_dict.keys() instead.

5. core/behavioral/behavioral_metrics.py — accepts a ucns_store parameter, but behavioral_window.compute_behavioral_windows never passes it. The UCNS marker comparison path is dead in normal pipeline flow. Either wire it up or remove the parameter.

6. canon/loader.py::CanonLoader() — re-reads and re-parses JSON on every construction. compute_round and compute_transcript create a fresh CanonLoader() when canon=None, causing per-round or per-transcript file reads at scale. Use a module-level singleton or require callers to pass the loader in.

[erinepshovel-code]

Two additional items for the sweep:

7. core/behavioral/structural_proxies.py::detect_list_growth_pattern — defined but never called anywhere. behavioral_metrics.py hardcodes O_list = 0 instead of calling it. Dead function.

8. backend/src/edcmbone/closed_tokens.py — imports unit_obj and multiply from ucns_v04, but neither is referenced in the module body. Dead imports (in addition to the installability issue tracked in [S1] Backend package not installable — closed_tokens.py imports ucns_v04 from repo root #46).

---

Generated by Claude Code