Fixes #1040 and fixes #1045.

By calling str on all stash elements we ensure they don't raise an error.
Worse case, soemthing like `<Element 'div' at 0x000001B2DAE94900>` gets
inserted into the output. However, with the override in the md_in_html
extension, we actually serialize and reinsert the original HTML. Worse case,
an HTML block which should be parsed as Markdown gets skipped by the
extension (`<div markdown="block"></div>` gets inserting into the output).

The tricky part is testing as there should be no known cases where this
ever occurs. Therefore, we forefully pass an etree Element directly to
the method in the test. That said, as #1040 is unresolved at this point,
I have tested locally with a real existing case and it works well.

Related to #1040.

* Ensure unclosed script tags are parsed correctly by providing a workaround for https://bugs.python.org/issue41989.
* Avoid cdata_mode outside of HTML blocks, such as in inline code spans.

Fixes #1036.

If pygments is installed and the version doesn't match the expected version.
then any relevant tests will fail. To avoid failing tests due to different
output by pygments, those tests will be skipped. The pygments tox env
sets the `PYGMENTS_VERSION environment variable, so that env will always
run those tests against the expected version.

* Pygments specific tests now only run when the pygments version installed
  matches the expected version. That version is defined in an environment
  variable (PYGMENTS_VERSION) in the 'pygments' tox env (see #1030).
* When the Python lib tidylib is installed but the underlying c lib is not,
  the relevant tests are now skipped rather than fail. This matches the
  behavior when the Python lib is not installed. The tox envs are now useful
  on systems which don't have the c lib installed.

* All non-language classes should always be assigned to the pre tag.
* The language identifying class should never be included with the
general list of classes.

Fixes #1032

Closes #1030

A second function, `slugify_unicode` was added rather than changing the existing function so as to maintain backward compatibility. While an `encoding` parameter was added to the `slugify` function, we can't expect existing third party functions to accept a third parameter. Therefore, the two parameter API was preserved with this change.

The HTML parser has been completely replaced. The new HTML parser is built on Python's html.parser.HTMLParser, which alleviates various bugs and simplifies maintenance of the code.

The md_in_html extension has been rebuilt on the new HTML Parser, which drastically simplifies it. Note that raw HTML elements with a markdown attribute defined are now converted to ElementTree Elements and are rendered by the serializer. Various bugs have been fixed.

Link reference parsing, abbreviation reference parsing and footnote reference parsing has all been moved from preprocessors to blockprocessors, which allows them to be nested within other block level elements. Specifically, this change was necessary to maintain the current behavior in the rebuilt md_in_html extension. A few random edge-case bugs (see the included tests) were resolved in the process.

Closes #595, closes #780, closes #830 and closes #1012.

Fixes #918.