[Enesuygurs]

Hard breaks are added with a plain regex over the span text, so a line ending inside an inline HTML tag gets a <br /> too. With break-on-newline this happens to any tag whose attributes are split across lines:

markdown2.markdown('Look at <img src="cat.png"\n     alt="A cat"> this.', extras=['break-on-newline'])

<p>Look at <img src="cat.png"<br />
     alt="A cat"> this.</p>

The browser ends the <img> at the <br /> and shows alt="A cat"> as text. Inline links with a wrapped title are affected the same way (title="multi<br />\nline title"), and so are the default two-space breaks when the spaces are inside a tag.

The core hard-break substitution and the breaks extra now share a _do_hard_breaks helper. It swaps tags that contain a newline for placeholders, applies the break regex, and then restores the tags. It uses _sorta_html_tokenize_re, the same approach smarty-pants has used since #735. Single-line tags aren't touched, so the ul/ol/li lookahead and everything outside tags behave as before. The extra pass only runs when the text contains both < and a newline.

I ran every .text file in the test dirs through master and this branch with breaks in each mode, with and without safe mode, plus a run combined with other extras. Apart from the new cases, the output is identical.

I added the break_on_newline_html_tags and hard_break_html_tags test cases, which both fail on master. make testone and make testredos pass.