Enrich MarkdownAST with origin of data

[fingolfin]

This now came up in multiple issues: it would be good if we kept track of where the MarkdownAST nodes come from (which file / lines).

This could be done by adding custom node types; or by keeping a global Dict (WeakKeyDic) which maps these nodes to location info. The goal would be to enable better error messages pointing to the precise origin of the problem

I am in particular thinking about tracking the origin of code blocks, so that could mean we replace MarkdownAST.CodeBlock by our own version that tracks the origin.

But also for broken [bla](@ref) links this is important; those could come from a docstring in some .jl file and can be hard to track down (see also #2296)

Tracking docstring origins might also be helpful for dealing with "syntax errors" in those (see #2821)