SHA-512 accelerated by x86 (alternative)

[karel-m]

This is an alternative to #731

@MarekKnapek I have taken your sha512_x86.c from #731 and:

• applied the aligned/unaligned access patch (_mm256_load_si256/_mm256_store_si256 >> _mm256_loadu_si256/_mm256_storeu_si256)
• made a few other tweaks needed to fit the latest sha1/sha2 changes
• basically following approach used in sha256_x86 in current develop as much as possible

Could you please test whether this PR works as expected?

[MarekKnapek]

Works fine on Linux. Tested with clang22 and (simulated) Nova Lake CPU. Works fine with clang22 UBSan. Works fine on MSVC Windows, but I needed to adjust an ifdef, should work since Visual Studio 2022. Tested with Visual Studio 2026. Tested with (simulated) Lunar Lake CPU.

Edit: Also tested with clang-cl, works fine. Sadly, my CPU simulator (Intel SDE) does not simulate the xgetbv instruction on Windows, so I manually disabled it for the tests.

[karel-m]

@sjaeckel this PR is IMO ready to be merged into develop

[sjaeckel]

LGTM