Recompute stack base in forked postmaster children.

This is for the benefit of running postgres under the rr
debugger. When using rr signal handlers running while a syscall is
active use an alternative stack. As e.g. bgworkers are started from
within signal handlers, the forked backend then has a different stack
base than postmaster. Previously that subsequently lead to those
processes triggering spurious "stack depth limit exceeded" errors.

Discussion: https://postgr.es/m/20200327182217.ubrrl32lyfhxfwk5@alap3.anarazel.de

src/backend/postmaster/postmaster.c

@@ -4900,11 +4900,6 @@ SubPostmasterMain(int argc, char *argv[])
 	/* Close the postmaster's sockets (as soon as we know them) */
 	ClosePostmasterPorts(strcmp(argv[1], "--forklog") == 0);
 
-	/*
-	 * Set reference point for stack-depth checking
-	 */
-	set_stack_base();
-
 	/*
 	 * Set up memory area for GSS information. Mirrors the code in ConnCreate
 	 * for the non-exec case.

src/backend/utils/init/miscinit.c

@@ -92,6 +92,15 @@ InitPostmasterChild(void)
 {
 	IsUnderPostmaster = true;	/* we are a postmaster subprocess now */
 
+	/*
+	 * Set reference point for stack-depth checking. We re-do that even in the
+	 * !EXEC_BACKEND case, because there are some edge cases where processes
+	 * are started with an alternative stack (e.g. starting bgworkers when
+	 * running postgres using the rr debugger, as bgworkers are launched from
+	 * signal handlers).
+	 */
+	set_stack_base();
+
 	InitProcessGlobals();
 
 	/*