P0009: Add design justification for static->dynamic and dynamic->static mdspan assignment

should they allow static -> dynamic conversion (seems fine)
should they allow dynamic -> static (note that this have UB)

Both of these cases are important. Static-to-dynamic conversion is "type
erasure," for when an implementation doesn't want to specialize for
compile-time extents. Dynamic-to-static conversion is "specialization" (the
opposite of type erasure), for when an implementation is willing to pay the
compile-time cost of specializing. A common use case for the latter is in
applications that solve partial differential equations through a variety of
different discretization methods. Many methods result in arrays with known
fixed extents, but making all of them compile-time constants might cost too
much compilation time, rather than just optimizing the most common methods.

Start by locating the P0009 proposal and its section on static-to-dynamic and dynamic-to-static mdspan assignment. Read the surrounding design discussion and existing comments, then add justification for both type erasure and specialization use cases; done means the proposal explains why each conversion matters.