gh-131316: handle NULL values returned by HACL* functions

[picnixz]

In a follow-up PR I'd like to move all UPDATE functions in a single helper. The construction is always the same and it would help the HMAC PR as well (I've done it out there).

More generally, I'd like to have the same construction for MD5/SHA1/SHA2/SHA3 as they really share common code. I don't know how much I can change the code so that it looks better, which is why I didn't change anything here.

We had a discussion on the blake2 PR concerning the cosmetics of blake2module and I think we can definitely improve how it's currently structured, because the functions become quite ugly. But that'd be another issue.

cc @msprotz@chris-eibl

• Issue: Handle error scenarios of HACL* functions #131316

[chris-eibl]

assert((Py_ssize_t)(LEN) <= (Py_ssize_t)UINT32_MAX);

Most probably you've already found out: Py_ssize_t is only signed 32bit in 32bit platforms (same for Windows and Linux).
Hence, (Py_ssize_t)UINT32_MAX gets negative and the assert always fires:
https://godbolt.org/z/1K1v5r6db

[picnixz]

Hence, (Py_ssize_t)UINT32_MAX gets negative and the assert always fires:

Yeah... that's what I observed afterwards... so I just removed the assertion (can't fail if the assertion isn't here 😈)

[gpshead]

FWIW it is quite reasonable for these update loops to use a size lower than UINT32_MAX, by the time data gets that large the computation already takes a long time so just looping over 1-2gb chunks up to INT32_MAX to fit within our smallest signed size constraint is fine.