On our server we set the crawler to create a new WACZ file after crawling 10GB of content to avoid these issues. Multiple WACZs can then be loaded with a replay.json file in ReplayWeb.page.
If you open the WACZ files, extract the WARCs and piece them together using py-wacz into a bunch of smaller files, that may get around some of the issues here? As for exactly what they are, unsure.
I think I’d file the uploading issue as a bug report?? If files that big are out of scope, we should impose a size limit.