Split TLS implementation into its own translation unit.

Split TLS implementation into its own translation unit.

SkTLS has it's own header separate from SkThread, and having SkThread own
the platform implementation of SkTLS is problematic with Chromium. The
simplest way to clean this up is to put the implementation in its own set
of files, where it is also more easily found.

R=robertphillips@google.com

Committed: https://code.google.com/p/skia/source/detail?r=10105

[bungeman-skia, 2013-07-15 21:29:09]

From looking into crbug.com/258499 it appeared that
<chromium>/src/skia/ext/SkThread_chrome.cc should implement
sk_atomic_load_memory_order_relaxed or some such thing which can call
NoBarrier_load. However, it because apparent that, for the last year at least,
we have not been using SkThread_chrome.cc at all. In attempting to bring this
file up to date simply to test, it became apparent that the fact that hiding
Skia's TLS implementation inside SkThread's implementation is going to cause
problems here, because Chromium does not expose the low level hooks we require.
Splitting the implementation of TLS out allows for us to proceed using
SkThread_chrome.cc for atomics and mutexes.

[robertphillips, 2013-07-16 14:44:12]

lgtm + 2 questions/suggestions

https://codereview.chromium.org/19240007/diff/1/src/ports/SkTLS_none.cpp
File src/ports/SkTLS_none.cpp (right):

https://codereview.chromium.org/19240007/diff/1/src/ports/SkTLS_none.cpp#newc...
src/ports/SkTLS_none.cpp:9: 
initialize to NULL?

https://codereview.chromium.org/19240007/diff/1/src/ports/SkTLS_win.cpp
File src/ports/SkTLS_win.cpp (right):

https://codereview.chromium.org/19240007/diff/1/src/ports/SkTLS_win.cpp#newco...
src/ports/SkTLS_win.cpp:10: 
intialize to false?

[bungeman-skia, 2013-07-16 14:59:28]

Committed patchset #2 manually as r10105 (presubmit successful).