FnRefs are a problem

[danakj]

I finally wrote a memory bug in subdoc and it was painful and i miss not having memory bugs.

The bug was using a fnref as an lvalue, which then is left with a dangling ref to a lambda. This is forbidden in the docs but I forgot. We need that clang-tidy to prevent it, or change FnRefs to heap alloc and be more like DynFnRef (like DynError) to do type-erasure to the heap.

[danakj]

This is resolved, as FnRefs are gone. We now have the concepts and type-erased DynFn which can be referenced or boxed.

Fixed by #356