Rewrite & cleanup HACL*-based extension modules

[picnixz]

I recently rewrote the _blake2 module. I want to the same for other modules where I can align the naming of functions across the different modules. The reason why I'm doing this is because it's becoming harder to make "similar" changes everywhere. In general, if I need to change something in MD5, then I also need to change it in SHA1/SHA2/SHA3, and it's easier if the code looks similar elsewhere.

There are some places that need to be updated because they are dead code, e.g.:

/*[clinic input]
module _sha2
class SHA256Type "SHA256object *" "&PyType_Type"
class SHA512Type "SHA512object *" "&PyType_Type"
[clinic start generated code]*/

should be

/*[clinic input]
module _sha2
class SHA256Type "SHA256object *" "clinic_state()->sha256_type"
class SHA512Type "SHA512object *" "clinic_state()->sha512_type"
[clinic start generated code]*/

Linked PRs

• gh-135532: use _Py_strhex in _md5.md5.hexdigest #135742
• gh-135532: optimize calls to PyMem_Malloc in SHAKE digest computation #135744
• gh-135532: cleanup clinic module directives for cryptographic modules #135822
• gh-135532: use defining_class for copying BLAKE-2 and SHA-3 objects #135838
• gh-135532: fortify DEBUG checks when fetching HACL*-based module state #135844

Bug fixes (3.14+)

• gh-135532: update python_hacl_namespaces.h #135741
• [3.14] gh-135532: update Modules/_hacl/python_hacl_namespaces.h (GH-135741) #135745

Backported PRs (HMAC-only)

• gh-135532: simplify handling of HACL* errors in _hmac #135740
• [3.14] gh-135532: simplify handling of HACL* errors in _hmac (GH-135740) #135743

Abandoned

• gh-135532: module-wide cleanup for HACL*-based extension modules #135536
• gh-135532: rewrite SHA-2 with macros #135821
• gh-135532: rewrite SHA-3 with macros picnixz/cpython#2

[picnixz]

I've rewritten SHA-2 and SHA-3 modules with generic macros but I'm torn between the now cleaner modules and the fact that I entirely rewrote them. SHA-2 is smaller but SHA-3 is larger.

[picnixz]

Offline I found other stuff that need to be updated independently:

• SHA-3 types in Argument Clinic code.
• MD-5 digests computations don't use _Py_strhex.
• Useless memory allocation for 0-length SHAKE hashes (we can switch conditions checks)
• Usage of unsigned char where we could use uint8_t: HACL* code expects an uint8_t so we can consistently use it.
• Inconsistentcies in naming classes: we have _sha2.SHA256Type vs _sha3.sha3_224. Note SHA-2 doesn't expose a __new__ method but SHA-3 does. MD-5/SHA-1/SHA-1 don't have a tp_new slot but SHA-3 and BLAKE-2 do. I don't know which one is the best, but I don't think we should have both.