Currently we do a deep filesystem search and only return results when that has finished. I think we should return results from the current directory immediately. This is particularly important for the file system crawler search provider.

I think it is probably fine to continue to do a full tracker search since that should always be fast.

One way to do this could be to instantiate two versions of the simple search, one that searches the current directory and one that searches everything else. We could then add a "loaded" signal or similar that indicates to the search directory that results may be shown but that searches are still in progress.

The two instances of the simple provider could be differentiated by either settings a property on one of them or sending one a different query. Seems like using a property may be cleaner since it will also allow us to differentiate the instance more directly when results are returned by just checking that property.