Unit test improvements to fully validate the SkImage contract

Fixed bug in SkSurface_Gpu to make the surface receive the new copy when copy
on write happens.
Committed: https://code.google.com/p/skia/source/detail?r=8622

[reed1, 2013-03-28 17:28:16]

I cannot judge the change to the copy-on-write part (where we now call
createCompatibleDevice), brian will have to evaluate. Is that change orthogonal
to the new (public) apis about imagedata?

[bsalomon, 2013-03-28 17:33:44]

On 2013/03/28 17:28:16, reed1 wrote:
> I cannot judge the change to the copy-on-write part (where we now call
> createCompatibleDevice), brian will have to evaluate. Is that change
orthogonal
> to the new (public) apis about imagedata?

That part lgtm

[Justin Novosad, 2013-03-28 17:41:47]

On 2013/03/28 17:28:16, reed1 wrote:
> I cannot judge the change to the copy-on-write part (where we now call
> createCompatibleDevice), brian will have to evaluate. Is that change
orthogonal
> to the new (public) apis about imagedata?

The imagedata part helped me write the test in a clean way.

[Justin Novosad, 2013-03-28 17:43:26]

On 2013/03/28 17:41:47, junov wrote:
> The imagedata part helped me write the test in a clean way.
Also, I will need this to bootstrap the chrome compositor.

[Justin Novosad, 2013-03-28 18:37:25]

Patch.  Removed API change.

[bsalomon, 2013-03-28 19:16:43]

On 2013/03/28 18:37:25, junov wrote:
> Patch.  Removed API change.

lgtm

[reed1, 2013-03-28 20:57:13]

Lets land the fix for sure.

I still don't agree with the premise of the tests, and would like to talk about
them separately (from the fix).

[Justin Novosad, 2013-04-11 17:53:28]

On 2013/03/28 20:57:13, reed1 wrote:
> Lets land the fix for sure.
> 
> I still don't agree with the premise of the tests, and would like to talk
about
> them separately (from the fix).

Done.  Landed without the test as r8622