I don't recall exactly (figures) but I think there was some game that depended on zero-size point sprites. I want to say Everquest 2, but it's as close to a random guess as it gets... Also I think there might have been some "real" Windows driver (maybe Intel?) where the test failed. So it's probably no huge deal if we drop these tests.
Otherwise, absolutely correct that we should handle the case explicitly ourselves (although it looks quite awkward to discard in the fragment shader depending on point size). AFAIU the GL spec doesn't specify what's supposed to happen in that case, we were mostly lucky with implementation-dependent behavior.
If there was maybe a Windows game that depends on it, and it works on Windows on at least NVidia and AMD, then yeah, I'm inclined to think we should handle it correctly.
I have no particular insight about the depth bias tests.
Oops, I didn't notice there was another part of these commits :-/
I don't think we should be removing that broken() without a better understanding of why that test fails.
OK I'll mark them as todo instead, until they can be fixed properly.