[NerlnetApp] Critical bugs in distributed running

[leondavi]

Fixes in this PR:

• There was no a multipart implementation from cowboy to cowboy http communication - Fixed
• Flask is very limited in terms of serving many connections at once. Therefore, batches can't be sent from cowboy in a manner of transaction per batch. I've implemented an ets that collects all of results data in main server and send it as a single chunk of data.
• Sending json to devices was never tested on multiple devices and there were bugs there that I fixed
• Due to flask issue that some of post messages get 200 ok but actually not handled I've added a
• Add ack from apiserver back to main server to improve the redundancy of the ack mechanism in case of bad connection to apiserver
• Fixed issue that caused sourceWaitingList to be empty and cut the flow of experiment. This is due to the fact that sources gets their source piece one by another. The fix is to send total count of sources to know how many are expected to take a part in experiment phase.