Second attempt to fix #1185 (forkProcess and -threaded)

Patch 2/2: first patch is to ghc

This time without dynamic linker hacks, instead I've expanded the
existing rts/Globals.c to cache more CAFs, specifically those in
GHC.Conc.  We were already using this trick for signal handlers, I
should have realised before.

It's still quite unsavoury, but we can do away with rts/Globals.c in
the future when we switch to a dynamically-linked GHCi.