fix/allow asyncio page splitting in nested event loops (#104)

We're now using asyncio for page split concurrency, but because the
client itself is not async, we need to manage our own event loop. This
complains if your environment already has a running event loop. For
instance, setting `split_pdf_page=True` in a jupyter cell will give you
`RuntimeError: This event loop is already running`.

Turns out there's a simple library to allow for nested event loops. We
just apply the monkeypatch in split_pdf_hook.py and the error goes away.

To verify, you'll need to run `pip install -e .` to install the local
version of the client. Run `make run-jupyter` and open up the sample
notebook in `_jupyter/`. Try making a request with page splitting
enabled and you'll see the above error. Then, check out this branch,
install locally again, restart your jupyter kernel, and the error is
fixed.

Author: awalker4

setup.py

@@ -37,6 +37,7 @@
         "jsonpath-python>=1.0.6",
         "marshmallow>=3.19.0",
         "mypy-extensions>=1.0.0",
+        "nest-asyncio>=1.6.0",
         "packaging>=23.1",
         "pypdf>=4.0",
         "python-dateutil>=2.8.2",

src/unstructured_client/_hooks/custom/split_pdf_hook.py

@@ -8,6 +8,7 @@
 from typing import Any, Coroutine, Optional, Tuple, Union
 
 import httpx
+import nest_asyncio
 import requests
 from pypdf import PdfReader
 from requests_toolbelt.multipart.decoder import MultipartDecoder
@@ -40,6 +41,8 @@
 MIN_PAGES_PER_SPLIT = 2
 MAX_PAGES_PER_SPLIT = 20
 
+nest_asyncio.apply()
+
 
 async def run_tasks(tasks):
     return await asyncio.gather(*tasks)