So that problem is in DownloadClient#inflate on line 561. UTF-8 is a multi-byte character encoding and the buffer might contain half the bytes of a character, which will cause those \uFFFDs in the decoded string. DownloadClient#streamToString does it the right way, so you can work around this issue by removing the Accept-Encoding header from the request for now (#inflate should probably be fixed; I'm surprised this hasn't come up earlier).
However, now I'm getting a stack overflow during code generation.
However, now I'm getting a stack overflow during code generation.