Fix build with NOSSE2

[puleglot]

Disable AESNI and SHANI code that depends on SSE2 availability if NOSSE2 is enabled.

[idrassi]

Thanks for picking this up. The unmodified patch's 32 bit crypto libraries passed EncryptionTest::TestAll on my machine and under QEMU Pentium III model with SSE2 unavailable. There's useful progress here, but I found some issues that need addressing before merging:

1. The amd64 build still fails: Disabling SHA removes the indirect wchar_t declaration from <immintrin.h>. Camellia.c then fails when including Common/Crypto.h. Adding <stddef.h> before that include fixes the build; the algorithm tests passed.

2. Please keep runtime AES acceleration: -mno-sse2 already disables AES intrinsics. Aes_hw_cpu.asm is built separately, so CRYPTOPP_DISABLE_AESNI unnecessarily disables AES acceleration and the userspace Camellia fast path. Removing it passed the tests after the header fix. Please also keep object selection and callers consistent for NOSSE2=1 NOAESNI=1.

3. The SHA guards don't agree: With GCC's -msha -mno-sse2, __SHA__ still enables the TrySHA256 call while CRYPTOPP_DISABLE_SHANI removes its implementation, causing a link failure. Please align the guards, preserving the NOASM and UEFI exclusions from #1896.

[idrassi]

I have pushed a small followup addressing the points from my review:

• Add the missing header for wchar_t in the amd64 build.
• Preserve runtime AES/Camellia acceleration and handle NOSSE2=1 NOAESNI=1 consistently.
• Align the SHA probe guards with its implementation, retaining the NOASM and UEFI exclusions.

Tested under QEMU Pentium III without SSE2.

Merging. Thank you for your contribution.