Investigate why parsed-code-block doesn't work with caption by default

[RastislavTuranyi]

In the default sphinx code-block directive, when the caption option is passed in, the default literal_block node is wrapped inside a container node. When this is replicated in this extension (for compatibility with the caption option), all the children of the parsed_code_block node get deleted/removed by sphinx somewhere between the return from the directive run method and the doctree-read event being emitted by sphinx.

This has been successfully circumvented by creating a deepcopy and keeping a reference to it inside the parsed_code_block node. The reference to the deepcopy is not used anywhere, but its mere existence prevents the children from being removed. Setting the children to the deepcopy:

        custom_node.extend(deepcopy(text_nodes))

does not help and neither does setting the children after the wrapping in container:

        if caption:
            custom_node = container_wrapper(self, custom_node, caption)
            custom_node.children[1].children = []
            custom_node.children[1].extend(text_nodes)

I am mystified why the children keep getting deleted/removed and completely baffled why the hack used works, but it does work and the sphinx/docutils engine is too complex to spend time trying to untagle the causes of this issue at this thime. Nevertheless, it would be good to do so and create a proper solution, or at least understand why the solution works.