cli: parallelize formatter calls with thread pool

reformat_draft now uses concurrent.futures.ThreadPoolExecutor with a
default of 8 workers. OpenAI SDK calls are synchronous but network-
bound, so threads release the GIL during I/O and give real speedup.
ThreadPoolExecutor.map preserves input order regardless of completion
order, so output is deterministic.

Empirical: HML draft (134 entries) went from ~25 min serial to 1m55s
with concurrency=12. ~12x speedup; further increases hit OpenAI rate
limits.

CLI gains a --concurrency flag (default 8) for tuning per draft size /
rate limit headroom. concurrency=1 forces serial execution for
debugging. New test asserts that order is preserved when concurrent
calls finish out of input order (uses a sleep-by-index fake formatter).

src/cmos/cli.py

@@ -9,6 +9,7 @@ from __future__ import annotations
 
 import argparse
 import sys
+from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
 from typing import Callable
 
@@ -17,16 +18,36 @@ from cmos.parser import split_bibliography
 
 FormatterFn = Callable[[str], str]
 
+DEFAULT_CONCURRENCY = 8
 
-def reformat_draft(text: str, formatter: FormatterFn | None = None) -> str:
+
+def reformat_draft(
+    text: str,
+    formatter: FormatterFn | None = None,
+    concurrency: int = DEFAULT_CONCURRENCY,
+) -> str:
     """Rewrite the bibliography section of ``text`` using ``formatter``.
 
     The formatter takes a single messy entry and returns the CMOS 18 form.
     If omitted, the real OpenAI-backed formatter is used.
+
+    Entries are reformatted concurrently using a thread pool of size
+    ``concurrency``. OpenAI SDK calls are synchronous but network-bound, so
+    threads give real speedup. ``ThreadPoolExecutor.map`` preserves input
+    order regardless of which calls finish first, so the output is
+    deterministic.
+
+    Set ``concurrency=1`` for sequential execution (useful for
+    deterministic debugging or when hitting rate limits).
     """
     fmt = formatter or format_bibliography_entry
     parsed = split_bibliography(text)
+
+    if concurrency <= 1 or len(parsed.entries) <= 1:
         rewritten = [fmt(entry) for entry in parsed.entries]
+    else:
+        with ThreadPoolExecutor(max_workers=concurrency) as pool:
+            rewritten = list(pool.map(fmt, parsed.entries))
 
     pieces: list[str] = []
     if parsed.before:
@@ -46,11 +67,20 @@ def main(argv: list[str] | None = None) -> int:
 
     fmt = sub.add_parser("format", help="Reformat a draft's bibliography to CMOS 18.")
     fmt.add_argument("path", type=Path, help="Path to a markdown draft.")
+    fmt.add_argument(
+        "--concurrency",
+        type=int,
+        default=DEFAULT_CONCURRENCY,
+        help=(
+            "Number of parallel formatter calls (default: "
+            f"{DEFAULT_CONCURRENCY}). Set to 1 for serial execution."
+        ),
+    )
 
     args = ap.parse_args(argv)
 
     if args.command == "format":
         text = args.path.read_text(encoding="utf-8")
-        sys.stdout.write(reformat_draft(text))
+        sys.stdout.write(reformat_draft(text, concurrency=args.concurrency))
         return 0
     return 2

tests/test_cli.py

@@ -66,3 +66,38 @@ def test_reformat_draft_bibliography_heading_preserved():
     draft = "## Bibliography\n\nentry.\n"
     output = reformat_draft(draft, formatter=_fake_formatter)
     assert "## Bibliography" in output
+
+
+def test_reformat_draft_preserves_order_under_concurrency():
+    # With concurrent execution the formatter is called on all entries in
+    # parallel; the CLI must reassemble them in input order regardless of
+    # which call finishes first. This test uses a fake formatter that
+    # sleeps based on the entry content so earlier entries finish AFTER
+    # later ones if order were naively tied to completion.
+    import time
+
+    def slow_fake(messy: str) -> str:
+        # Entries with lower index sleep longer so they finish last.
+        n = int(messy.split()[-1])
+        time.sleep(0.05 * (5 - n))
+        return f"FORMATTED({messy})"
+
+    draft = """\
+## Bibliography
+
+entry 0
+entry 1
+entry 2
+entry 3
+entry 4
+"""
+    output = reformat_draft(draft, formatter=slow_fake, concurrency=4)
+    # Entries must appear in input order.
+    lines = [l for l in output.splitlines() if l.startswith("FORMATTED(")]
+    assert lines == [
+        "FORMATTED(entry 0)",
+        "FORMATTED(entry 1)",
+        "FORMATTED(entry 2)",
+        "FORMATTED(entry 3)",
+        "FORMATTED(entry 4)",
+    ]