OSX: Remove some nsnotificationcenter observers more reliably.

OSX: Remove some nsnotificationcenter observers more reliably.

There are some crashers which show up as an objc_msgSend when posting
a notification.  So I went around reviewing our observers looking for
places where it's possible that we weren't unregistering correctly.
These may have already been correct, but this change would seem to
make unregistration more reliable.

[Scott Hess - ex-Googler, 2009-08-26 00:13:46]

This isn't really intended to explicitly fix anything, just good hygiene.

[rohitrao (ping after 24h), 2009-08-27 16:30:19]

http://codereview.chromium.org/174480/diff/1/4
File chrome/browser/autocomplete/autocomplete_popup_view_mac.mm (left):

http://codereview.chromium.org/174480/diff/1/4#oldcode309
Line 309: object:[field_ window]];
This move feels a little strange.  Is the matrix_target_ dealloc'ed every time
the popup is dismissed?  Or are we going to end up receiving window resize
notifications even when the popup is closed?

[Scott Hess - ex-Googler, 2009-08-28 20:04:40]

http://codereview.chromium.org/174480/diff/1/4
File chrome/browser/autocomplete/autocomplete_popup_view_mac.mm (left):

http://codereview.chromium.org/174480/diff/1/4#oldcode309
Line 309: object:[field_ window]];
On 2009/08/27 16:30:19, rohitrao wrote:
> This move feels a little strange.  Is the matrix_target_ dealloc'ed every time
> the popup is dismissed?  Or are we going to end up receiving window resize
> notifications even when the popup is closed?

Ooh, good catch, the observer is longer lived than any individual observee. 
I'll put this back (while leaving the -dealloc remove).

I don't think this would have ended up observing all windows.  Best case, it
would have just accumulated a bunch of unused observers.  Worst case, it was
storing them in an NSDictionary and things would have crashed!

[rohitrao (ping after 24h), 2009-08-31 21:02:10]

LGTM