This is a great example of including native(-ish) code in Lambda! I’ve got a couple of comments.
First, there’s an official Docker image for Amazon Linux, and the Lambda team have said they’ll put a tag on it for the current version in use by Lambda.
Second, I don’t think building (or pip installing) code inside an Amazon Linux container for Lambda is a hack at all! In fact, it should probably be a common tactic for mature teams, as you never know when a Python package will have native requirements. Similarly, having a base Docker image that’s Amazon Linux with LLVM seems pretty useful too!