Misc: CMake tidy-up, replace non-constant shift immediates in GSVector

[stenzek]

Description of Changes

Works around the requirement for the offsetof parameter to be constant.

Moves GSVector shift immediates over to a template parameter, since you can't use non-constant shift immediates for NEON, so this needs to happen anyway. And, less importantly, apparently GCC 13 doesn't like it.

Also removes the __m128 overloads - it's too easy to mistake for a variable shift (which doesn't exist in SSE4), instead it takes the shift amount from the lowest 32-bits.

Removes a bunch of redundant checks from the CMake scripts, and uses standardized platform names.

Adds a warning in caps for anyone trying to build with GCC (we only support Clang/MSVC).

Rationale behind Changes

Janitor part 3?
Closes #10500, although GCC 13 is still broken on Ubuntu 23.10.

Suggested Testing Steps

Make sure software renderer still works.
Note to self: do a dump run before merging.

[Ziemas]

Compiles on gcc 13.2.1 for me, software renderer seems to work. 👍

[stenzek]

It doesn't build on Ubuntu 23.10, or Fedora 39:

In function ‘__m128i _mm_srli_si128(__m128i, int)’,
    inlined from ‘GSVector4i GSVector4i::srl() const [with int i = 4]’ at /home/user/dev/pcsx2/pcsx2/GS/GSVector4i.h:694:41,
    inlined from ‘void GSState::GIFPackedRegHandlerSTQRGBAXYZ2(const GIFPackedReg*, u32) [with unsigned int prim = 5; bool auto_flush = false; bool index_swap = false]’ at /home/user/dev/pcsx2/pcsx2/GS/GSState.cpp:690:20:
/usr/lib/gcc/x86_64-linux-gnu/13/include/emmintrin.h:1229:10: error: the last argument must be an 8-bit immediate
 1229 |   return (__m128i)__builtin_ia32_psrldqi128 (__A, __N * 8);
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Could be a bug, I'm not sure, need to dig further.
But not really a blocker since we don't support gcc anyway.