Scaffold CMOS 18 reformatter: harness, linter, parser, formatter, CLI

Initial phase-1 baseline of the karpathy/autoresearch-style loop.
The formatter module is the inner-loop artifact; parser and linter
are infra. The linter carries a LINTER_VERSION hash (v0.2.0) that
will force a re-baseline on any rule change.

Components:
- harness/diff.py: case-sensitive field-level substring diff
- harness/score.py: three-axis scoring (field, linter, canary exact)
- src/cmos/linter.py: 9 CMOS 18 structural rules, each Purdue/CMOS cited
- src/cmos/parser.py: locate ## Bibliography section, split entries
- src/cmos/formatter.py: prompt + OpenAI call with caller injection
- src/cmos/cli.py: cmos format path/to/draft.md
- scripts/run_loop.py: loop runner with --fake mode for no-API runs
- exemplars/: 3 canary seed exemplars (book, journal w/DOI, web),
  sourced from chicagomanualofstyle.org quick guide

Tests: 48 passing. Fake-mode baseline scalar = 0.000 on the 3 seed
exemplars (identity caller fails the linter on every rule). This is
the floor the real GPT-5 formatter needs to improve from.

tests/test_diff.py

@@ -0,0 +1,53 @@
+"""Tests for harness/diff.py — field-level structured diff.
+
+The diff checks, for each (field, expected_value) in a canonical record,
+whether the candidate formatted string contains that value as a case-sensitive
+substring. This is intentionally simpler than parsing — the linter catches
+structural issues; the diff catches "did the right facts survive?"
+"""
+
+from harness.diff import field_diff
+
+
+def test_all_fields_present_in_candidate_match():
+    canonical = {
+        "author": "Smith, Jane",
+        "title": "The History of Nothing",
+        "publisher": "University of Chicago Press",
+        "year": 2023,
+    }
+    candidate = "Smith, Jane. *The History of Nothing*. University of Chicago Press, 2023."
+    result = field_diff(candidate, canonical)
+    assert result.matched == ["author", "title", "publisher", "year"]
+    assert result.missing == []
+    assert result.rate == 1.0
+
+
+def test_missing_publisher_is_reported():
+    canonical = {
+        "author": "Smith, Jane",
+        "title": "The History of Nothing",
+        "publisher": "University of Chicago Press",
+        "year": 2023,
+    }
+    # Publisher dropped.
+    candidate = "Smith, Jane. *The History of Nothing*. 2023."
+    result = field_diff(candidate, canonical)
+    assert "publisher" in result.missing
+    assert "author" in result.matched
+    assert result.rate == 0.75
+
+
+def test_title_case_mismatch_counts_as_missing():
+    # CMOS 18 requires headline-style caps. If the formatter emits sentence
+    # case, the title should not match its canonical headline-case form.
+    canonical = {"title": "The History of Nothing"}
+    candidate = "Smith, Jane. *The history of nothing*. 2023."
+    result = field_diff(candidate, canonical)
+    assert result.missing == ["title"]
+    assert result.rate == 0.0
+
+
+def test_empty_canonical_is_rate_one():
+    # Degenerate case — no fields to check means nothing is wrong.
+    assert field_diff("whatever", {}).rate == 1.0