Use a bitmask for SVG render object types

Use a bitmask for SVG render object types

This patch changes the virtual isSVG*() functions into regular
functions that check a bitmask (via the new virtual "getSVGType"). This
should have no impact on performance. For reference, a similar pattern
is used in Node.h.

This change adds RenderObject::isSVG() with no new virtual calls. In a
future patch, this new function will let us to remove the
O(height of tree) setNeedsBoundariesUpdate call in
RenderObject::willBeRemovedFromTree :)

Additionally, this change opens the door for some useful cleanups. For
example, RenderSVGResource has RenderSVGResourceMask which can be
folded into RenderObject::SVGType. Similarly, there is no
isRenderSVGModelObject() function today but this will be trivial to
add after this patch.

For reference, I put together a diagram of the SVG render tree hierarchy:
https://docs.google.com/a/chromium.org/drawings/d/1nayJX5XcclYj_-blbVuakVNjVWdRAzC5i5s6xBeHfzA/edit

[pdr., 2013-12-16 06:51:21]

This is a bit experimental and a lot of work for just isSVG(). What do you
think?

[pdr., 2013-12-17 00:43:47]

On 2013/12/16 06:51:21, pdr wrote:
> This is a bit experimental and a lot of work for just isSVG(). What do you
> think?

Ojan has convinced me we should go the route of adding virtual bool isSVG()
instead.

This complexity is fragile and, if we need it in the future, we can rip out the
virtuals and land this patch.

[pdr., 2013-12-17 05:12:49]

On 2013/12/17 00:43:47, pdr wrote:
> On 2013/12/16 06:51:21, pdr wrote:
> > This is a bit experimental and a lot of work for just isSVG(). What do you
> > think?
> 
> Ojan has convinced me we should go the route of adding virtual bool isSVG()
> instead.
> 
> This complexity is fragile and, if we need it in the future, we can rip out
the
> virtuals and land this patch.

Closing this issue. New and improved approach:
https://codereview.chromium.org/108203005/