etl/delegate: fix accident creation of a delegate to an rvalue delegate when copying/assigning from delegate with mismatching signature

[VladimirP1]

Original problem, how to reproduce:

#include <etl/delegate.h>

namespace ctx {
struct os {};
struct any {
  any() = default;
  any(os) {}
};
} // namespace ctx

struct Test {
    etl::delegate<void(ctx::os)> GetCb() {
      return etl::delegate<void(ctx::any)>::create<Test,
      &Test::CbAnyCtx>(
          *this);
    }

private:
  void CbAnyCtx(ctx::any ctx) {}
};

int main() {
  Test().GetCb()(ctx::os{});
  etl::delegate<void(void)> f = []() {};
  return 0;
}

This compiles and does not look suspicious at the first glance but results in a segmentation fault at runtime.

What happens here:

1. A delegate is created etl::delegate<void(ctx::any ctx)>::create<Test, &Test::CbAnyCtx>(*this);
2. But we need a etl::delegate<void(ctx::os ctx)>, the compiler selects the constructor from const TLambda&.
3. Note that etl::is_same<etl::delegate<TReturn(TParams...)>, TLambda>::value is false because the signatures of the delegates do not match
4. The newly created delegate is returned and the first delegate is destroyed. Now the second delegate holds a dangling pointer

This PR attempts to fix this issue by using a reliable method for checking if a functor is a delegate

[jwellbelove]

Thanks, I'll take a look at that when I get back from Spain later this week.

[jwellbelove]

Does class delegate have any use apart from being a base class 'tag'?
It's mainly a style thing, but when I've done something like this before I've usually just created a simple 'tag' class.

struct delegate_tag {};

class delegate<void> sort of looks like it should do something useful as a delegate type.

[VladimirP1]

No other use, I simply did not want to pollute the namespace with additional classes

[VladimirP1]

Should I replace delegate<void> with a tag class?

[jwellbelove]

Yes, I think so.

[jwellbelove]

It may be good to make a traits class.

template
etl::is_delegate;

[jwellbelove]

Just thought of another addition.

etl::is_delegate_v definition.
For C++17 only.

[jwellbelove]

I should be able to process this pull request tomorrow afternoon.