runtime: avoid read overrun in heapdump

Start the stack a few words below the actual top, so that
if something tries to read goexit's caller PC from the stack,
it won't fault on a bad memory address.
Today, heapdump does that.
Maybe tomorrow, traceback or something else will do that.
Make it not a bug.

TBR=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/136450043

src/runtime/proc.c

@@ -1047,6 +1047,7 @@ runtime·newextram(void)
 	gp = runtime·malg(4096);
 	gp->sched.pc = (uintptr)runtime·goexit;
 	gp->sched.sp = gp->stack.hi;
+	gp->sched.sp -= 4*sizeof(uintreg); // extra space in case of reads slightly beyond frame
 	gp->sched.lr = 0;
 	gp->sched.g = gp;
 	gp->syscallpc = gp->sched.pc;
@@ -2229,6 +2230,7 @@ runtime·newproc1(FuncVal *fn, byte *argp, int32 narg, int32 nret, void *callerp
 		runtime·throw("newproc1: new g is not Gdead");
 
 	sp = (byte*)newg->stack.hi;
+	sp -= 4*sizeof(uintreg); // extra space in case of reads slightly beyond frame
 	sp -= siz;
 	runtime·memmove(sp, argp, narg);
 	if(thechar == '5') {