With every single change in a document the corresponding IndexItem
elements were extracted and put into an vector. The locator filters
then used these lists to iterate over and filter them.

This change removes that extraction, and the filtering now directly
iterates over the IndexItem elements with a callback.

The exception is the current document filter, because it also queries
for all declarations. Adding this to the model would result in a higher
memory usage, while that information is only used by this filter.

Change-Id: Ibe445cc11e9f68b5d807348fd46c7cac4aff4c85
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@digia.com>