previously, when consumer want to connect to remote host, the engine internally created 2 channels to the remote host. one for the consumer itself, and the other to move remotable data required for the consumer in background.

this was done so that each consumer can have its own isolated connections from others. but later discovered that it can cause too much allocations since something like code lens can cause remote host to have 10+ more concurrent requests and syncing all those remotable data at once can cause too high allocation rate.

so, in RTM we throttled on how many can sync remotable data at the same time, but left 2 channels design as it was.

this change makes connection to have only 1 channel, and remotable data channel to be shared by all connections. since there is time delay to make connection to remote host, this should reduce delay required per each connection.