Allow for a preconfigured number of Tendermint node failures

We have a long-term requirement coming from [GoT](https://github.com/interchainio/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.

Start by tracing how tm-load-test handles Tendermint node connection failures in standalone mode and master/slave mode. Identify the configuration entry point and connection retry path, then verify that both modes tolerate the configured number of failures and keep retrying failed endpoints without aborting the load test.

Written by the indexing model from the issue text.