Add support for cache in resources

[cyppan]

I think it's better to implement caching as a dedicated feature instead of composing a DataSource.

Once the oplog and a tailable cursor has been implemented, it's pretty easy to add a long-lived memory cache which would be an atom referencing a map of $resource-$id keys to document futures.
This way requests are de-deduplicated de facto.
In the case of querying several ids at one time, every id can point to a future extracted from the multi-id running future.

A good simplification IMO:

• when a query is made against ids directly, use the cache directly when possible (otherwise feed it with results).
• when a query is made using a find specification, get the ordered ids from that query using the datasource directly, and then fetch the corresponding documents from the cache if they exists, else feed it.