Fix memoryview on 64-bit PowerPC for floats: workaround a GCC bug
when a 64-bit double is casted to a 32-bit float followed by
memcpy().

The memoryview pack_single() function is miscompiled by GCC 8.2.1 on
PowerPC 64. Pseudo-code:

    void
    f (double d, char *target)
    {
      float f = d;
      __builtin_memcpy (target, &f, sizeof (f));
    }

When optimizations are enabled (-Og or -O1 at least), GCC rounds
towards zero rather than rounding to nearest:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88892

Use the volatile keyword to prevent the optimization which uses the
wrong rounding mode.