Policies and contextual bindings are broken.

[gitmodimo]

I have been experimenting with uml_dumper and some weird connections we made in dependency graph.
I made a deep dive into how it works and I believe that uml_dumper is fundamentally broken.
I will go into details in uml_dumper case but I believe this is also true for all policies and contextual bindings due to the same mechanism beeing used.

In uml_dumper parent-child relationship is based on std::vector<const std::type_info*> v = {&typeid(nullptr)};.
When first element is constructed it gets added to vector N=T::arity::value times. IIUC this is due to next N policies invocations would presumably be due to creation of current elements dependencies.
However this assumption is wrong in three major cases:

1. A singleton is being used the second time - then singletons' dependencies are not created again, but the singleton is pushed to std::vector<const std::type_info*> v nevertheless. I tried fixing it with conditionally adding singletons only when they are not already on the list, but this seems to be flaky.
2. Dynamic binding is used - then T::arity::value does not represent number of dependencies. AFAIK dynamic bindings may have dynamic number of dependencies.
3. Multiple Bindings - T::arity::value would need to represent number of elements in vector, but it doesn't. Also uml_dumper does not compile when multiple bindings are used.