Since this affects only tests, they could easily change the relevant scripts to download a list of test videos at runtime. I bet the RIAA github-scrapers would not "see" it. Just serve statically a rot13-encoded list of URLs from pastebin or something, and Bob's your uncle.
Based on what I saw in past discussions, I'm pretty sure that the takedown was not a run-of-the-mill scraper-based takedown (it makes no sense to be taken down just for linking to videos which, at best, is what any scrapers would have seen in the original test code). It was very much an intentional, manual one with actual lawyers behind it.
There are multiple sides to "defending" a project like this. One of them is avoiding to trip the run-of-the-mill scrapers. The takedown was serious but we don't know what triggered the lawyers' attention in the first place. IMHO a simple runtime obfuscation would remove that particular attack vector, once coupled with some plausible deniability (i.e. deleting all downloaded data once tested). At that point, YTDL is still on RIAA's generic shitlist (which will require other mitigations to survive) but at least doesn't get flagged every week by a scraper.
