This causes the artificial reductions in the Connect REST request timeout to be more isolated. Specifically, they now only take place in the tests that need them (instead of any tests that happen to be running after the reduction has taken place and before it has been reset), and they are only performed for the requests that are expected to time out, before being immediately reset. This should help reduce spurious test failures (especially in slow environments like Jenkins) for all Connect integration tests that interact with the REST API, not just the BlockingConnectorTest test suite.

Reviewers: Bruno Cadonna <cadonna@apache.org>