SkNEW_PLACEMENT(Ganesh, memory management)

Addresses http://code.google.com/p/skia/issues/detail?id=687 (nee http://code.google.com/p/skia/issues/detail?id=598) by wrapping SkNEW macros around all the new calls in Ganesh.

Adds support for placement new to SkNEW macro family defined in SkPostConfig.h.

[reed1, 2012-07-09 18:28:53]

How does your new macro fit in with the utilities in SkTemplatesPriv.h?

[TomH, 2012-07-09 18:32:55]

I didn't know about those!

I don't like the look of them, though. #ifdef SK_BUILD_FOR_WIN32 throws away
placement new? Ganesh has been using placement new on Win32 without any
problems, as far as I know.

[reed1, 2012-07-09 18:39:10]

Lets fix 'em if they're broken, and decide if all placement macros should be in
the same file

[TomH, 2012-07-09 19:35:32]

Patch set 2 uses SkTemplatesPriv.

At the moment I have four problems with it:

1. turns off placement new on Windows
2. requires include of private header file in include/
3. calls new somewhere else, defeating our intent in addressing the original
bugs of giving our users a single change point to intercept new and provide
their own memory manager
4. fails the assert at the very end of SkPictureFlat

+Justin to see if he has any idea why we're breaking that assert, since it came
in with his checksum change in r4378?

[TomH, 2012-07-09 20:20:10]

After verbal LGTM from bsalomon@ and reed@, reverted to patchset 1 and committed
as r4492.

Mike will investigate #1.
#2 can be worked around by adding a cpp file to implement that chunk of
GrProgramStageFactory, which probably doesn't need to be inlined.
#3 can be hacked around with documentation, although that weakens the power of
SkUserConfig.h
#4 turns out to be my source tree being out-of-date, missing r4478.