Add thread safety to Component updater singleton

In some tests the component updater singleton might be called first from the IO thread which is not expected.

This is observed today.

BUG=none
TBR=jam

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=232551

[cpu_(ooo_6.6-7.5), 2013-11-02 01:57:38]

Committed patchset #2 manually as r232551.

[Nico, 2013-11-02 05:01:37]

I'm not sure if this is the right fix, but maybe I just lack background.
http://build.chromium.org/p/chromium.chromiumos/builders/Linux%20ChromiumOS%2...
started failing browser_tests with this assertion, but not just for the
component updater, also with several other stacks.

(e.g.
http://build.chromium.org/p/chromium.chromiumos/builders/Linux%20ChromiumOS%2...
but lots of others too)

I'd guess that some other change messed up threads somehow.

[cpu_(ooo_6.6-7.5), 2013-11-02 05:16:40]

I can sort of explain, the component updater singleton needs to be created in
the UI thread. The changes of bug 307193 exposed the singleton to the IO thread.
Which is safe for one method of it but not for component updater creation.

Then some tests call the IO thread component updater singleton ahead of the UI
thread.