Allow for a preconfigured number of Tendermint node failures

[thanethomson]

We have a long-term requirement coming from GoT where experiments will need to be conducted under load where some nodes might fail. At present, the default behaviour of tm-load-test is to fail the entire load test completely.

We need tm-load-test to:

1. be configurable in the number of Tendermint node failures it can tolerate, and
2. have the ability to keep retrying to connect to failed Tendermint node endpoints.

This functionality needs to apply to both master/slave mode and standalone mode.