limit check pushed inside loop

[Henry Cejtin]

I  mailed  Matt,  but  then  I  realized  that it didn't go out to MLton: The
example he gave, conjecturing that the coalesced limit checks  violate  safe-
for-space,  does  not  violate  it.  The guarentee is just that for any fixed
program the O(space) is not changed.  Moving limit checks earlier, as done in
MLton, can only increase the space required by O(1), so it is fine.

A  worse case, which is still allowed under safe-for-space, is the conversion
of some tail calls to non-tail calls.  This never happens in a loop (or  else
we  would  not  be  safe-for-space) but can multiply the number of live stack
frames by as much as the number of lambda expressions in the program.  Again,
this is still ok as far as space safety.