Fix premature HTML-escaping of inline code

[mentalisttraceur]

Fixes #372 by moving HTML-escaping of codespan from the parser into the HTML renderer.

Passes all tests, and test_ast_output is enhanced to cover this case. (I just ran pytest, are there other tests or any linters I should run?)

Before (parse to AST wrongly shows raw codespan content as &, which is a bug for use-cases that never render to HTML - for comparison, & outside of codespan doesn't get escaped):

>>> import mistune
>>> md = mistune.Markdown()
>>> md('`&`')
[{'type': 'paragraph', 'children': [{'type': 'codespan', 'raw': '&'}]}]

After (pure parse correctly shows raw code content is &):

>>> import mistune
>>> md = mistune.Markdown()
>>> md('`&`')
[{'type': 'paragraph', 'children': [{'type': 'codespan', 'raw': '&'}]}]

After (still works when rendering to HTML):

>>> md.renderer = mistune.HTMLRenderer()
>>> md('`&`')
'<p><code>&amp;</code></p>\n'

Use-case example: I have a script which publishes files written in Markdown as Tumblr posts. It walks through Mistune's AST, generating Tumblr's "NPF". I need the raw contents, because Tumblr takes the raw text and does its own HTML rendering+escaping server-side. Currently, my script has to have extra handling for codespan tokens to unescape the characters escaped by mistune.util.escape.

P.S. Special thanks to @torokati44 for finding the exact line in the parser that was causing the problem.