When barebox is invoked out of the TF-A v2.1, it's started with
sp, r0, r1, r2 all equal to zero. To use the new RAM size calculating
stm32mp1_barebox_entry, we need to have a stack to handle spillage.

Add a stm32mp_cpu_lowlevel_init wrapper around arm_cpu_lowlevel_init,
which additionally configures a 64 byte stack after the end of the
barebox binary. This should be enough to help us through the RAM size
calculation. If not, compression will fail because of data corruption
and stack size can be increased as necessary.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>