math: extend non-glibc guard to Clang and add exp10 guard

[kjg0724]

Follow-up to #1398, addressing two remaining issues reported in #1396.

1. Clang on non-AArch64 targets

The previous fix guarded __builtin_roundeven for GCC 10+ only, based
on the assumption that Clang always lowers it to an inline
llvm.roundeven.* intrinsic. That is only true on AArch64 (FRINTN).
On x86 without SSE4.1, powerpc, sparc, i386, and others, Clang emits
a roundeven()/roundevenf() libm call — same failure mode as GCC.

Fix: drop the compiler-specific check and apply the glibc >= 2.25 guard
to all compilers. Non-glibc platforms fall through to the existing
portable inline fallback.

2. __builtin_exp10 / __builtin_exp10f

exp10() is a GNU extension present only in glibc. On musl and other
non-glibc libcs the symbol is absent, producing the same class of link
error. The existing fallback (simde_math_pow(10.0, v)) is correct;
it just needs to be activated on non-glibc platforms.

Closes #1396.

[brad0]

arm64$ cat /tmp/test.c
double foo1(double x) {
    return __builtin_roundeven(x);
}

float foo2(float x) {
    return __builtin_roundevenf(x);
}

long foo3(long x) {
    return __builtin_roundevenl(x);
}
arm64$ /usr/local/bin/clang-19 -target aarch64-unknown-openbsd -O2 -c test.c -o test.o
arm64$ nm -u test.o
arm64$ /usr/local/bin/clang-19 -target powerpc-unknown-openbsd -O2 -c test.c -o test.o
arm64$ nm -u test.o
roundeven
roundevenf
arm64$ /usr/local/bin/clang-19 -target powerpc64-unknown-openbsd -O2 -c test.c -o test.o
arm64$ nm -u test.o
.TOC.
roundeven
roundevenf
arm64$ /usr/local/bin/clang-19 -target mips64-unknown-openbsd -O2 -c test.c -o test.o
arm64$ nm -u test.o
roundeven
roundevenf
arm64$ /usr/local/bin/clang-21 -target amd64-unknown-openbsd -O2 -c test.c -o test.o
arm64$ nm -u test.o
roundeven
roundevenf

[brad0]

Thanks. I'll take this for a spin and test on the various archs OpenBSD / FreeBSD.

[brad0]

Build tested and tests run on OpenBSD/ FreeBSD amd64 with GCC and Clang. Build tests and tests run on OpenBSD aarch64 with Clang.

This also closes #1014

[mr-c]

Thank you @kjg0724 !

[mr-c]

@brad0 @kjg0724 PR(s) with additional CI configuration (non-Github Actions preferred) that would have caught some of these issues would be very welcome.

@brad0 thank you for the manual testing!