Progressive index import

[tchoutri]

At the moment, each index import job tries to import the whole index. This is inefficient, as only the delta between the saved index and the current index could be imported.

We should explore ways to timestamp the files that have been already imported, compute the delta, and feed it to the import job.