arch

gcc-7 optimizes the byte-wise accesses of get_unaligned_le32() into
word-wise accesses if the 32-bit integer output_len is declared as
external. This panics then the bootloader since we don't have the
unaligned access fault trap handler installed during boot time.

Avoid this optimization by declaring output_len as byte-aligned and thus
unbreak the bootloader code.

Additionally, compile the boot code optimized for size.

Signed-off-by: Helge Deller <deller@gmx.de>