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_score.py

@@ -0,0 +1,116 @@
+"""Tests for harness/score.py — the frozen scoring function.
+
+The score function takes a list of exemplars and a formatter callable,
+produces per-exemplar results, and aggregates a scalar for the loop.
+"""
+
+from pathlib import Path
+
+from harness.score import Exemplar, load_exemplars, score
+
+
+def identity_formatter(messy: str) -> str:
+    return messy
+
+
+def perfect_formatter_for(fixtures: dict) -> callable:
+    """Return a formatter that looks up the expected output by messy input."""
+
+    def _fmt(messy: str) -> str:
+        return fixtures[messy]
+
+    return _fmt
+
+
+BOOK = Exemplar(
+    name="book_single_author",
+    source="toy fixture",
+    type="book",
+    tags=["book"],
+    canary=False,
+    messy_input="smith, jane. the history of nothing. u of chicago press, 2023.",
+    canonical={
+        "author": "Smith, Jane",
+        "title": "The History of Nothing",
+        "publisher": "University of Chicago Press",
+        "year": 2023,
+    },
+    expected_bibliography="Smith, Jane. *The History of Nothing*. University of Chicago Press, 2023.",
+)
+
+WEB = Exemplar(
+    name="web_page",
+    source="toy fixture",
+    type="web",
+    tags=["web"],
+    canary=False,
+    messy_input="modern language association. mla style center. 2024.",
+    canonical={
+        "author": "Modern Language Association",
+        "title": "MLA Style Center",
+        "year": 2024,
+    },
+    expected_bibliography='Modern Language Association. "MLA Style Center." 2024. https://style.mla.org/.',
+)
+
+
+def test_identity_formatter_scores_below_one():
+    # Identity returns the messy input unchanged, so canonical fields should
+    # mostly NOT appear in the output.
+    result = score([BOOK, WEB], identity_formatter)
+    assert result.scalar < 1.0
+    assert len(result.per_exemplar) == 2
+
+
+def test_perfect_formatter_scores_one():
+    fixtures = {
+        BOOK.messy_input: BOOK.expected_bibliography,
+        WEB.messy_input: WEB.expected_bibliography,
+    }
+    result = score([BOOK, WEB], perfect_formatter_for(fixtures))
+    assert result.scalar == 1.0
+    assert result.field_match_rate == 1.0
+
+
+def test_load_exemplars_reads_toml_files(tmp_path: Path):
+    (tmp_path / "example.toml").write_text("""
+source = "toy"
+type = "book"
+tags = ["book"]
+canary = true
+messy_input = "smith jane, nothing, 2023"
+expected_bibliography = "Smith, Jane. *Nothing*. Publisher, 2023."
+
+[canonical]
+author = "Smith, Jane"
+year = 2023
+""".strip())
+    loaded = load_exemplars(tmp_path)
+    assert len(loaded) == 1
+    ex = loaded[0]
+    assert ex.name == "example"
+    assert ex.canary is True
+    assert ex.canonical["author"] == "Smith, Jane"
+    assert ex.canonical["year"] == 2023
+
+
+def test_canary_exact_match_tracked_separately():
+    canary_book = Exemplar(
+        name="book_canary",
+        source="toy",
+        type="book",
+        tags=["book"],
+        canary=True,
+        messy_input=BOOK.messy_input,
+        canonical=BOOK.canonical,
+        expected_bibliography=BOOK.expected_bibliography,
+    )
+
+    # Formatter produces correct fields but with a trailing space — fields
+    # match, exact-string doesn't.
+    def nearly(_messy: str) -> str:
+        return BOOK.expected_bibliography + " "
+
+    result = score([canary_book], nearly)
+    assert result.field_match_rate == 1.0
+    assert result.canary_exact_match_rate == 0.0