It does not interact properly with RLIMIT_STACK, and has a few other
problems as noted in PR 260303. Let's disable it for now until a better
solution is reached.
I suspect that we should stop copying psstrings into the stack, and
instead create a dedicated mapping for them. Then, map the full stack
at some bounded, randomized offset below this mapping, and let
RLIMIT_STACK apply to this mapping as usual, without any knowledge of
the stack gap.