handling the case of multiple top-level headers

[choldgraf]

I ran into a weird bug - if you have a page with multiple top-level headers, then the in-page TOC logic breaks down.

This is less-common (though possible) with rST, but consider in markdown a page like this:

My title
========

# My header 1

Content 1

# My header 2

Content 2

In this case, Sphinx will parse all of the headers at the same, top level. Our in-page TOC code assumes that the only top-level header is the title node, and that all other headers are underneath it.

This PR does the following:

• Check if there is more than 1 entry in the TOC object. If so, it means there is more than 1 header at the top level.
• Assume that the non-title top-level headers are the "true" headers for the document
• In the Jinja step, take the non-title top-level headers, and append them to the title's top-level header, so that we can render

The one awkward thing is that if there are other 2nd and 3rd level headers in-between the next 1st-level header, they will be missed in the TOC. Something like this:

My title
========

## My second level

# My header 1

Content 1

# My header 2

Content 2

The TOC would only show "My header 1" and "My header 2".

But I think that's probably the behavior we'd want to encourage?

Lemme know if you'd like me to explain a bit more...I think the longer-term solution will be to make the "get_page_toc" function a little bit less-hacky, but this seems like a reasonable patch for now

[jorisvandenbossche]

@choldgraf Can you rebase this? Or do you directly want to do a different solution as described in #197 ?

[choldgraf]

closing in favor of #199