Search works by using sd::outliner::OutlinerContainer to iterate over
all text objects of the document. We used to switch to each and every
object, and only then search in it. In large presentations this means
the number of invalidations during search was dependent on the number of
slides between the current slide and the first match.

Fix this by not calling sd::Outliner::SetObject() (which would call
sd::Outliner::SetPage()) right after finding a text object, only later
when we know it has matching content.

The result is that the number of invalidations is not O(n) but O(1) till
we find the first match.

Change-Id: I29a11c8737a7e1db6a247eb98617d12495c8bb41