Don't apply smarty-pants inside inline HTML tags

[LuisFigueroaG]

Fixes #150

With the smarty-pants extra on, the quote and dash substitutions ran over the whole span text, inline HTML tags included. Quotes inside attributes were turned into entities and broke the tag:

markdown2.markdown('<img src="/path/to/img.png" alt="Alt Text">', extras=["smarty-pants"])

<p><img src="/path/to/img.png&#8221; alt="Alt Text"></p>

Inline HTML comments broke too, because -- became an en dash: <!-- note --> came out as <!&#8211; note &#8211;>.

The footnote case from the original report was already handled by a special case in SmartyPants.run that changed class="footnote-ref” back, but other tags still had the problem.

What I changed: SmartyPants.run now swaps the inline tags and comments for hash placeholders first, using the existing _sorta_html_tokenize_re. Then it does the substitutions and puts the tags back. Since the placeholders aren't whitespace, the text around a tag looks the same to the quote regexes as it did before, so output outside tags doesn't change. With tags skipped, the footnote special case isn't needed any more, so I removed it. footnotes_smarty-pants still passes.

To check that nothing else changed, I ran every .text file in the test dirs through master and through this branch with smarty-pants on (alone and with footnotes, fenced-code-blocks and tables). The only differences were inside tags (src/width/title attributes and <!-- --> comments), plus the randomly encoded mailto links, which are different on every run anyway.

Tests:

• New case tm-cases/smarty_pants_html_attributes_issue150 covers an <img> with double-quoted attributes, an <a> with a single-quoted title, quotes inside a <span>, and an inline comment containing --. It fails without the fix.
• make testone: 289 tests, OK (288 on master)
• pycodestyle lib/markdown2.py: 413 lines vs 415 on master (no new warnings)
• make testredos: 7 tests, OK