[Enesuygurs]

With the toc extra, a document whose first header isn't its highest-level one produces broken TOC HTML. For example, ## Overview, # Getting Started, ## Install gives:

<ul>
  <li><a href="#overview">Overview</a></li>
</ul></li>
<li><a href="#getting-started">Getting Started</a>
<ul>
  <li><a href="#install">Install</a></li>
</ul>

The outer list closes straight away, a stray </li> is left behind, and everything after it ends up outside any list.

A related problem shows up when a level is skipped and the document later returns to it. # A, ### B, ## C, ## D puts C next to A at the top level and nests D under C, although both are h2s under A.

calculate_toc_html compared the new header with the current list's level instead of its parent's, and it could close the outermost list. It now closes a nested list only when the header belongs at or above the parent's level; otherwise the header stays in the current list. The outermost list is never closed early, so a header above the first one's level becomes another top-level entry. Properly nested documents give the same output as before, and unlike #337 this doesn't add extra nested lists for skipped levels.

The nesting now matches Python-Markdown's toc extension: I compared 20,000 random header sequences against its toc_tokens, and all of them match and are well-formed.

I added the toc_shallower_headers test case, which fails on master. make testone and make testredos pass.