Do not simply rely on the compiler to check if we should link with the math library

[G4m4]

Fix #92: first look for the math library using standard CMake, then only link with it explicitly if not compiling for Windows nor Mac.

[marton78]

Two blocking issues found:

---

Critical: MATHLIB uninitialized when PFFFT_DISABLE_LINK_WITH_M=ON on Unix

In the new code, when MATH_LIBRARY AND NOT WIN32 AND NOT APPLE is true but PFFFT_DISABLE_LINK_WITH_M is set, MATHLIB is never assigned. The opt-out branch needs an explicit set(MATHLIB "").

Suggested fix:

if(PFFFT_DISABLE_LINK_WITH_M)
  set(MATHLIB "")  # ← this line is missing
  message(STATUS "INFO: PFFFT_DISABLE_LINK_WITH_M set: will not link math lib m")
else()

---

Important: CI failures on build_win_mingw and build_win_msvc

The most likely cause is the change from set(MATHLIB "m") to set(MATHLIB ${MATH_LIBRARY}) — on Unix, find_library returns an absolute path (e.g. /usr/lib/libm.so), not just m. Passing an absolute path to target_link_libraries changes linker behavior and can break cross-compilation toolchains. The Windows CI failures need to be diagnosed with this in mind.

---

Minor issues

• find_library(MATH_LIBRARY m) runs unconditionally on all platforms including Windows, where it always returns NOTFOUND — better to guard with if(NOT WIN32 AND NOT APPLE)
• Missing status message when the opt-out flag is active (it was present in the original code)

---

The overall architecture is sound — using find_library plus platform guards is the right approach. These are fixable issues.