amd64: bring back asm bcmp, code shared with memcmp

[mjg]

Summary

Turns out clang converts "memcmp(foo, bar, len) == 0" and similar to bcmp calls, which makes it worth it to provide the dedicated routine.

[jhb]

Do probably want to change the recent change to depends-cleanup.sh to check for bcmp.c instead of bcmp.S (and update the date and git hash).

lib/libc/amd64/string/memcmp.S

150–168

Maybe s/Before we compute it/Before we compute mismatch,/ since the comment is now split up a bit and the antecedent of 'it' may not be as obvious now?

161

This is just the same as 'inc %eax' or 'add $1,%eax'?

I assume the simplification here is that you only need a non-zero value for bcmp() as it doesn't determine a tri-state like memcmp?

[emaste]

Agree with @jhb wrt the comment