After enabling logging of failed email parsing, I discovered that we are failing to parse a non trivial amount of emails. From what I can tell, this only applies to mboxes from Mailman 3, at least so far. An example of a problematic email header looks like this:

Subject:
 Integrated: 8371748: Remove the (empty) ThreadPoolExecutor.finalize() method

Header values are allowed to break lines, so long as the first character on the newline is whitespace. In this case, the whole value has been shifted down to the next line. The regular expression used to extract these headers needs a slight adjustment to handle this.

The same mail from the existing Mailman 2 server does not have this extra newline.