Proxy fixes for SkImageFilter

Fix recursive GPU processing for ImageFilters.  Plumb through the SkImageFilter::Proxy parameter to the GPU recursion path.  Extract DeviceImageFilterProxy from SkCanvas.cpp into its own .h, and rename it.

Background:  when a filter which implements onFilterImageGPU() calls a filter which doesn't, it passes a NULL Proxy.  This means that we switch to software rasterization for that filter, and never go back. The fix is to construct an SkDeviceImageFilterProxy from the SkGpuDevice, and pass that into onFilterImageGPU().

[reed1, 2012-09-27 20:28:49]

interface: lgtm

[Stephen White, 2012-09-27 21:19:06]

On 2012/09/27 20:28:49, reed1 wrote:
> interface: lgtm

Brian, would you like to take a look at the GPU bits?

[bsalomon, 2012-09-27 21:52:42]

looks perf to me. I mean perf good to me.... I mean lgtm.


On Thu, Sep 27, 2012 at 5:19 PM, <senorblanco@chromium.org> wrote:

> On 2012/09/27 20:28:49, reed1 wrote:
>
>> interface: lgtm
>>
>
> Brian, would you like to take a look at the GPU bits?
>
>
https://codereview.appspot.**com/6575059/<https://codereview.appspot.com/6575...
>

[Stephen White, 2012-09-27 23:52:22]

On 2012/09/27 21:52:42, bsalomon wrote:
> looks perf to me. I mean perf good to me.... I mean lgtm.

Perf you very perf.  I mean, thanks.  :)

Landed in https://code.google.com/p/skia/source/detail?r=5720.  Closing.
> 
> 
> On Thu, Sep 27, 2012 at 5:19 PM, <mailto:senorblanco@chromium.org> wrote:
> 
> > On 2012/09/27 20:28:49, reed1 wrote:
> >
> >> interface: lgtm
> >>
> >
> > Brian, would you like to take a look at the GPU bits?
> >
> >
>
https://codereview.appspot.**com/6575059/%3Chttps://codereview.appspot.com/65...>
> >