Minor leak fixes and valgrind suppressions for tests

Two leaks left: http://code.google.com/p/skia/issues/detail?id=530 and one in RefDict.

Several warnings about uninitialized data to go.

This suppression turns off all warnings about things that go wrong in the drivers. One or two of those might be able to be traced back to user code.

[epoger, 2012-03-15 18:04:41]

LGTM

http://codereview.appspot.com/5837046/diff/1/tests/BitmapCopyTest.cpp
File tests/BitmapCopyTest.cpp (right):

http://codereview.appspot.com/5837046/diff/1/tests/BitmapCopyTest.cpp#newcode597
tests/BitmapCopyTest.cpp:597: delete [] buf;
Is there not a good way to do this with a smart pointer instead of manual
delete?

[TomH, 2012-03-15 18:14:03]

Skia avoids smart pointers; you'll find a few RefPointers, but only in the PDF
code, which wasn't written by the main team.

The closest idiom we have is the raft of SkAuto* helpers; is this more pleasing?

[epoger, 2012-03-15 18:15:16]

On 2012/03/15 18:14:03, TomH wrote:
> Skia avoids smart pointers; you'll find a few RefPointers, but only in the PDF
> code, which wasn't written by the main team.
> 
> The closest idiom we have is the raft of SkAuto* helpers; is this more
pleasing?

Sure, I meant "smart pointers" only in the most generic sense.  Really, anything
that will cause the memory to be released upon leaving the scope of that method
is fine with me. :-)

[TomH, 2012-03-15 18:17:00]

Closed with r3406.