Fix portable is[inf|nan|_out compilation on older Linux

By wrapping a potentially non-compliant `isinf`/`isnan` implementations into a lambda with a defined return type

Compiler should be able to optimize it out into direct function call, see https://godbolt.org/z/bqYGd47Mx

Author: malfet

kernels/portable/cpu/op_isnan.cpp

@@ -15,8 +15,10 @@ namespace executor {
 namespace native {
 
 Tensor& isnan_out(RuntimeContext& ctx, const Tensor& in, Tensor& out) {
+  // Lambda is syntactic sugar needed to workaround compilation on some older non-compatible distros
+  // where isnan is not defined for double, but either for float or for long double
   return internal::unary_ufunc_realhb_to_bool(
-      static_cast<bool (*)(double)>(std::isnan), ctx, in, out);
+      [](double x) -> bool { return std::isnan(x); }, ctx, in, out);
 }
 
 } // namespace native