[Phase 3][Colocation] Use single RPC for joins

[ndeodhar]

Jira Link: [DB-389](https://yugabyte.atlassian.net/browse/DB-389)
Since colocated tables are stored on the same tablet, we can optimize join queries to use a single RPC to read data instead of multiple, separate RPCs.
This is similar to the optimization that we have done for index scans.

[azapedia]

Hi guys, thanks for the great work on YugabyteDB, want to ask if this is already addressed. I have tested the colocated table with simple and complex joins. Seems like the performances in some cases are not far too different. And in some cases, it is worst than a non-colocated table. And do you have a query cache?