feat(parsing): LaTeX, Nougat, GROBID parsers with chunking

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

codex/parsing/nougat.py

@@ -0,0 +1,54 @@
+"""Nougat OCR integration.
+
+Converts PDF files to Mathpix Markdown (mmd) by calling a self-hosted
+Nougat HTTP server.  No network fetching of papers happens here — the
+caller is expected to pass a local PDF path.
+"""
+
+from __future__ import annotations
+
+import httpx
+import tenacity
+
+from codex.config import Settings
+
+
+def pdf_to_markdown(pdf_path: str, nougat_url: str | None = None) -> str:
+    """Convert a local PDF to Mathpix Markdown via the Nougat HTTP API.
+
+    Parameters
+    ----------
+    pdf_path:
+        Absolute (or relative) path to the PDF file on disk.
+    nougat_url:
+        Base URL of the Nougat server.  Defaults to ``Settings().nougat_url``.
+
+    Returns
+    -------
+    str
+        The raw ``.mmd`` text returned by the server.
+
+    Raises
+    ------
+    httpx.HTTPStatusError
+        If the server returns a non-2xx status code.
+    """
+    if nougat_url is None:
+        nougat_url = Settings().nougat_url
+
+    @tenacity.retry(
+        retry=tenacity.retry_if_exception_type(httpx.ConnectError),
+        stop=tenacity.stop_after_attempt(3),  # 1 original + 2 retries
+        wait=tenacity.wait_fixed(0),
+        reraise=True,
+    )
+    def _post() -> str:
+        with open(pdf_path, "rb") as fh, httpx.Client(timeout=120.0) as client:
+            response = client.post(
+                f"{nougat_url}/predict",
+                files={"file": (pdf_path, fh, "application/pdf")},
+            )
+        response.raise_for_status()
+        return response.text
+
+    return _post()