x86_64: Fix long double to int cast

x86_64-gen.c:
- gen_cvt_ftoi()
    Use TOK___fixxfdi for VT_LDOUBLE independently of the size of the
    target integer type, avoiding a wrong conversion to doube.

    This solves:

    int main(void) {
        long double x = 0.999999999999999999l;
        return (int)x;
    }

    Before for VT_INT a long double was first converted to double with
    rounding, possibly leading to a wrong result when truncating with
    cvttsd2si.

x86_64-gen.c

@@ -2121,18 +2121,15 @@ void gen_cvt_ftoi(int t)
     ft = vtop->type.t;
     bt = ft & VT_BTYPE;
     if (bt == VT_LDOUBLE) {
-	if (t != VT_INT) {
         vpush_helper_func(TOK___fixxfdi);
         vswap();
         gfunc_call(1);
         vpushi(0);
         vtop->r = REG_IRET;
+        if (t != VT_INT)
 	    vtop->r2 = REG_IRE2;
         return;
     }
-        gen_cvt_ftof(VT_DOUBLE);
-        bt = VT_DOUBLE;
-    }
 
     gv(RC_FLOAT);
     if (t != VT_INT)