This means we must downgrade to SHA256, as SHA3 is not in hashlib.

This breaks the build, as Kyber and Dilithium currently are not
properly namespaced. More pressingly, this is not a long-term
solution: because of the way Makefile dependencies work,
one failing test now prevents all others from running.

We probably want to change the testing setup not to rely so heavily
on the Makefile, and instead call a Python or shell script.
Related, we may want to reconsider the single scheme / all targets;
this is becoming quite a mess in the Makefile, already.

Also fixes some tidy issues

Not sure why my local clang-tidy doesn't complain

This also adds some code for canaries, and unifies crypto_{kem,sign}

The previous formatting did not immediately stabilize,
and the resulting formatting is somewhat arbitrary.
Manually inserting linebreaks seems to make more sense.

Separating this into principal and auxiliary submitters was
chosen by NIST. If we do not do the same thing, we would have to
decide how to order names. People seem to have opinions on ordering.