libc: minimal: add size optimized string functions

The current implementations of memcpy and memset are optimized for
performance and use a word based loop before the byte based loop.

Add a config option that skips the word based loop. This saves 120
bytes on the Cortex-M0+ which is worthwhile on small apps like a
bootloader.

Enable by default if SIZE_OPTIMIZATIONS is set.

Signed-off-by: Michael Hope <mlhx@google.com>