Use cursor instead of first counting new files and then getting them

[MaxiBoether]

The Storage gRPC service, when checking for new files, first gets the number of new files to allocate a vector of the correct size, and then basically executes the same query again, this time dumping the actual data into the vector. The reason is that soci expects the vector to have a sufficient size.

To solve this, we could try and use a cursor instead, since we know we will be getting fixed size chunks there. Then we avoid doing the same work twice.