Make compiletest look for exact matches in the arch table

[hanna-kruppe]

#47737 reported a problem with architecture names that are substrings of each other, but the problem is more general than powerpc. #48732 removed the powerpc annotations, but I argued afterwards, it would be better to improve compiletest such that it doesn't consider e.g. powerpc to be a powerpc64 target. In light of the arguments linked earlier and the additional fact that this keeps happening -- #48875 introduces the same problem for MIPS -- I really think we should revert #48732 and fix compiletest instead.

cc @nagisa @jcowgill @debris @alexcrichton

[alexcrichton]

Sure! Just needs someone to do the work :)

[jcowgill]

#48875 introduces the same problem for MIPS

I don't think so. The change in #48875 works because "mips64" is inserted above "mips" in the ARCH_TABLE list.

[hanna-kruppe]

Hm, I think you're right. The correctness of the code depending on such a subtlety is still pretty awful, though.