Improve datasources import

[sylvinus]

We currently have ad-hoc scripts in https://github.com/commonsearch/cosr-back/tree/master/scripts to import our various datasources, each of them creating a different rocksdb database.

This is not ideal and should be improved in the following ways:

• [x] Use only one database. We may use https://github.com/facebook/rocksdb/wiki/Merge-Operator to update the keys with each new datasource, though I'm not sure it is needed if we only import sequentially.
• [x] Move the import code inside the URLServer
• [x] Stream the data from the network instead of downloading files and then reading them.
• [x] Have a look at low-hanging fruits performance-wise
• [ ] Write a guide on how to add a new datasource. Let's try to have the bar for experimentation as low as possible.
• [ ] Investigate the complexity of migrating to gRPC / Protobufs. I had a prototype of this but went with mprpc for the sake of simplicity.
• [ ] Update the architecture schema to include the URLServer