caseFieldAccessors: fixing me softly

I just committed in r19145 a patch which once again goes after ticket
#1373, but from a position of enhanced understanding. Since I
subtly broke stuff the last time I went after that bug, this time it is
extremely carefully examined and very thoroughly tested, but I figured I
had better also mention it here. The breakage last time was (I'm 99%
sure) a result of calling fullNameString on Symbol -- this time I only
use machinery which I know has already been used.

Iulian: I had to make a change to SpecializeTypes. The logic you had in
there was necessary only because of the incorrect logic being used in
caseFieldAccessors, but once that logic was corrected, your workaround
turned into an anti-workaround because I could no longer see the
original fields. So now it's an anti-anti-workaround.