Problematic codegen with __builtin_isfpclass and vector types

``__builtin_isfpclass`` is specified to have support for vector builtins
https://clang.llvm.org/docs/LanguageExtensions.html#builtin-isfpclass

but the codegen can also end up wrong (x86, arm, ...)
or suboptimal (avx512) when using it

Examples
https://minimal.godbolt.org/z/59eb3vbGe

Start with the linked Godbolt example and the Clang Language Extensions documentation for __builtin_isfpclass. Compare generated code for the vector cases across the affected x86, Arm, and AVX-512 targets, then identify the relevant compiler codegen and test locations. Done means correct codegen on the affected targets and improved AVX-512 output, covered by regression tests.

Written by the indexing model from the issue text.