Move early malloc() to before arch_cpu_init()

Move early malloc() to before arch_cpu_init()

For some CPUs, having malloc() available very early is useful. There is no
reason to delay this since early malloc is allocated before board_init_f()
is called.

Move early malloc() init nearer to the start of the init sequence.

Signed-off-by: Simon Glass <sjg@chromium.org>