Create an experiment to study whether too many bindings cause NAT failure.

In this experiment, we're trying to study the impact to NAT by the number of bindings created in short intervals. The stunprober will send 1 ping to each server. There will be total |rounds| of probers. They will be grouped into batches. For example, we'll have the first N probers reporting the success rate of total recv over total send as the first batch, the next N as the 2nd batch. Each batch will report its own UMA. This way, we could observe whether the later batch has higher failure rate than the initial ones.

The list of important parameters will be
1. the interval b/w the pings
2. the batch size
3. the total rounds

This depends on ongoing CL https://codereview.webrtc.org/1422593002

BUG=549639

Review URL: https://codereview.chromium.org/1417663004

Cr-Commit-Position: refs/heads/master@{#357293}
7 files changed