Fix antisite generation in mixed-valence systems

[kavanase]

Context: SMTG-Bham/doped#96

Flagged by @alexsquires, for mixed-valence systems AntisiteGenerator gives X_X substitutions. Using sets instead of lists for determining the elements in the structure fixes this issue (see discussion in linked issue). Added a test for this behaviour too, and confirmed it fixes the downstream issue noted in doped

[kavanase]

Btw the (previously) failing test here is unrelated to this PR.
Seems to be some small numerical noise in the test_ccd.py results (due to some dependency update I guess?), have fixed this by loosening the relative tolerance to 1e-4 here which is perfectly good.

[kavanase]

The other failing test now in the notebooks is again unrelated, something to do with the Freysoldt correction with sxdefectalign

[jmmshn]

@kavanase, sorry for missing this!

Looking at it now.

[jmmshn]

Again sorry for the late reply. I have shifted focus to some functional materials at MS over the past few months but I'll try to tweak my github settings to I catch these alerts better.

Thanks for the update on this.
I think this is fine. I'm still a bit on the fence about using formal oxidation states during defect creation so decision making in this part of the code is pretty arbitrary.

I'll merge and release soon. @tschaume thanks for brining this to my attention!