main should behave like hashsum, so it must pass the GNU Test Suite.

I'm considering to implement xorsl (xorsum "lite"), which is targeted at embedded systems:

• Its API must be similar to BusyBox cksum and md5sum, so it'll have less features.
• It'll only support 2 lengths (to avoid dynamic allocation): 4B & 8B.
• All the buffers must be as small as possible.

lib should have a similar API to BLAKE3, without all of the crypto-specific stuff.