Issue #3462

The mode associated with a token is now the inner mode at
the time when the token is started. The htmlmixed and markdown/gfm
modes were fixed to return empty tokens during mode switches
when necessary to enforce this rule.

Issue #2919

Bumping them for every release just adds too much git
noise and history blowup.

Also configure the htmlmixed option to pass through xml-mode-specific
options to its inner xml mode.

Issue #2265

Also moves handling of electricChars to readInput, removing
the timeout kludge.

Issue #2029

Issue #1943

Closes #1271

Closes #1248

Issue #1242
Issue #1198

Issue #1209

Issue #1198

This allows the xml mode to return a style with space separated classes
and still work. I didn't use /\btag\b/ since that can match "tag-foo".

And move closetag over to it.

This makes the code that gets the XML state out of the mode actually
sound and extensible.

Issue #820

 - Background parsing now never goes past the visible part of the
   document.

 - We only store style information for lines that are actually visible
   (and then keep it cached).

 - When the document changes, background highlighting always
   re-highlights from the change to the end of the visible part.

 - Gets rid of compareState mode methods and the hairy heuristic that
   tried to simulate it when absent.

 - The onHighlightComplete callback was removed, since it no longer
   really applies -- the document is only fully parsed when scrolled
   to its end.

This should help preserve memory (a huge document will no longer
immediately have parser state and highlighting information built up
for all lines, but only for the parts that you look at), and removes
the pathological case when you, for example, are typing at the top of
a huge XML document, and opening or closing a tag causes a whole
re-highlight to cascade all the way to the bottom because the
compareState will detect a change.

Issue #688

Closes #751

This to make it possible to mix modes with and without compareStates
implementations.

Closes #431

Since v2 is the canonical CodeMirror at this point

So that I can put other doc-related stuff there without polluting the
root dir.

It's tiny anyway, no need to force people to include yet another file.

To help external code easily determine which mode a getTokenAt token is in