Reported as https://github.com/mongodb-labs/mongo-arrow/issues/371

"I accidentally ran aggregate_pandas_all without my schema, and kept getting OverflowError. I went through and checked all the Int32 fields, and they were all well within the range. Adding the schema fixed the problem. This means that at some point, the schema inference is deciding something is an Int32, then later it's getting a larger value. I tried to debug a little more so I could provide a minimal test case, but I can't write to our mongodb so it's tricky. I suspect inserting a small int and a large int would trigger this."

We need to verify the bug, add a test, and add a fix.