by changing the tag before html is parsed and restoring it before markdown parsing

Compressed with `uglifyjs lib/marked.js --comments="/Copyright/" --mangle --compress`

Used Marked v0.3.3 as of chjj/marked@2b5802f258c5e23e48366f2377fbb4c807f47658

see https://github.com/hakimel/reveal.js/issues/929#issuecomment-80738837

add global flag back in for whitespace removal, without it most of the content in example.html does not work #682

renamed markdown attributes for clarity and consistency: data-vertical -> data-separator-vertical, data-notes -> data-separator-notes

- on slides: .slide x=y
- on elements: .element x=y

Those are detected in html comment, in order to not be visible
if a markdown is edited in an html editor.

The first slide of a vertical stack see some data attributes ignored.

Mainly the data-transition one.

Repeat all data-attributes on the wrapping section element.

Ignore any other attributes (like 'id="xxx"'), in order to not
mess with internal links (by repeating twice an id).

By default, look for <!-- slide-attributes: xxxx -->.
Whatever 'xxx' is will be added to the section attributes.

You can define your own pattern with 'data-attributes'.
For instance 'data-attributes="^\s*?-- (.*?)$"':
that will be added to the options.

The 'attributes' section is removed from the content of the slide,
so the final markdown doesn't reflect them.
That also means you can add those attributes *anywhere* in the slide

But that allows for *any* attribute to be added for a specifc slide,
like:

- id="plan", for allowing internal link (issue #430)
- data-background="#ff0000"
- data-transition="fade"

You list those attributes on a single line, like

- (default):
` <!-- id="plan" data-background="#ff0000" -->`
- or, with an alternative data-attributes pattern:
` -- id="plan" data-background="#ff0000"`

Again, that line is remove from the final content.

By default '{\\\.\s*?([^}]+?)}'.

Allows to add attributes to element of an attribute which contains
the attribute pattern.

Works better, and the html elements get their attributes.

Extra text representing classes is detected and correctly removed.
Adding attributes isn't working yet.