random retry elapsed time when all nodes down

As part of Amadeus Couchbase client test campaign, we encountered some random run time on retry strategy while performing a set operation towards a mock that has been stopped.

Reproducer to implement in Lcb:

1/ Connection lost with mock

After connection and traffic, stop the mock then run an operation (ie. set).

Please find attached logs with 3 runs:

• the first run lasts 254492 ms
• the second 8742 ms
• the second 5720 ms

it can vary from 3s to 300s.

Some times ago, a fix had already been done on this ticket CCBC-948 The library might stuck in indefinite retry of configuration get when all nodes down, Following that, we have created this unit test.