Fix Scalameta tests for 2.12.8

[wiwa]

No description provided.

[wiwa]

It's worth noting that MyersDiff is essentially O(N^2) (at least?) where N ~ 17000 for 2.12 predef... (this is where it "hangs").

It also would run very slowly for 2.11 predef, but we skip it since the scalasigs are equal. Need to investigate if the slowdown is okay. And should probably print to console that diffing takes a while. Of course, the best case is that we can just make the scalasigs equal.

[wiwa]

Diffing algo uses random lookups: use arraylist rather than a List. Also, use jgit diff algo instead of myers (this alone is like 20x faster...).