The line calculation was ignoring paddingTop and scrolling.

Should help trigger #836, and is cleaner anyway.

Closes #840

When the editor is hidden, they can be null.

I don't know what they were thinking, but this, on
recent Opera versions, seems to give the correct
result.

Closes #10

Closes #745

Closes #754

Not linked yet. Beta comes out later this week.

It was accidentally calling global getSelection.

This broke right-click copy.

- Prevent "markdown_hrDashLong" test from failing when runHarness() is called manually.

- Render EM / STRONG within link text (closes #823).

- Add highlighting for images (closes #824).

They were completely broken due to really fragile use of html
selectors. Should work much better now in test framework.

Closes #796

Line widgets could make drag-select break before.

Closes #800

Since these no longer follow directly from scroller.clientHeight/Width.

Use these to ensure messages for current line are scrolled into view
in demo/widget.html.

Closes #802

Closes #808

Huge refactor of lib/codemirror.js. Most code is now top-level (in the
module wrapper function) rather than inside of the CodeMirror
constructor sharing a huge amount of state.

A CodeMirror instance now has (non-public) `display`, `view`, and
`options` properties that describe its state. The display holds the
editor's DOM structure and associated state, view represents a view on
a document, containing `doc`, `sel`, `overwrite`, `scrollTop`, and
`scrollLeft` properties. The `doc` itself contains its history.

(It is not yet possible to switch views and documents in and out of an
editor, but this does start clearing the way towards that.)

That's what you get when you test with the published version,
rather than the one you're actually editing.