Remove AVX-512 backend paths and improve ARM/MSVC SIMD detection

[falseywinchnet]

Motivation

• Remove the AVX-512-specific backend and 512-bit intrinsics so the kernel supports only scalar, 128-bit (SSE2/NEON), or AVX2+FMA paths.
• Make NEON detection and FMA detection more robust for MSVC/Windows ARM64 targets so Windows ARM64 builds pick the NEON or scalar paths correctly.
• Simplify documentation and internal notes so they reflect the supported SIMD backends and avoid describing AVX-512 experiments.

Description

• Stripped AVX-512 selection and all __m512 / _mm512* codepaths from src/detail/bruun_kernel.hpp, including conditional blocks that compiled 512-bit kernels.
• Added a BRUUN_HAS_FMA probe (__FMA__ || _M_FMA) and adjusted the AVX2 level test to require BRUUN_HAS_FMA, and updated 128-bit x86 and NEON detection to consider MSVC macros (_M_IX86_FP, _M_ARM64, _M_ARM64EC, __ARM_NEON__, and __ARM_NEON).
• Reworked conditional includes so immintrin.h and arm_neon.h are pulled only for the supported backends and guarded by the improved detection macros.
• Updated runtime backend naming, inline comments, and supporting docs/notes (documentation/python.md, src/detail/inverse_acceleration_notes.txt, src/detail/kernel_sequentiality_flow.md) to remove or reword AVX-512 references.

Testing

• Ran a full CMake build with auto-SIMD detection and tests via cmake -S . -B build -DBFFT_BUILD_TESTS=ON -DBFFT_BUILD_EXAMPLES=OFF && cmake --build build -j2 && ctest --test-dir build --output-on-failure, and all tests passed.
• Ran a no-auto-SIMD build and tests via cmake -S . -B build-no-auto -DBFFT_ENABLE_AUTO_SIMD=OFF -DBFFT_BUILD_TESTS=ON -DBFFT_BUILD_EXAMPLES=OFF && cmake --build build-no-auto -j2 && ctest --test-dir build-no-auto --output-on-failure, and all tests passed.
• Searched the source tree for AVX-512 tokens (rg -n "AVX-512|avx512|__AVX512|__m512|_mm512") and found no remaining AVX-512 occurrences.
• Performed a clang++ -target aarch64-pc-windows-msvc -fsyntax-only attempt to validate MSVC/Windows ARM target handling, which failed in this container due to missing Windows C++ standard library headers and therefore could not fully validate MSVC system headers here.

---

Codex Task