Turns out sometimes docid can be a string.

For small counts, this reported wrong number of documents (batch size
instead of the small count).

This allows to query large sets in batches and send the results
upstream.

Patch 2

See merge request !6

Add abstract, source, etc

See merge request !4

Servant update

See merge request !3

Made client take doc as a parameter so we can know precisely what compose it

search should always be called with `docid` and `label_s` if you don't want more
than this just use `basicSearch`