Add optional ping timeout

[preslavle]

Does your PR solve an issue?

The PR adds a optional and configurable ping timeout. Without ping timeout, some connections can get stuck indefinitely in rare circumstances, where a TCP connection is blackholed. I don't a reliable repro, but this is a clearly a problem that can happen, especially in cloud environments with live VM migrations, etc. This can affect the connection pool in two ways:

1. Ping performed during return_to_pool. Without a timeout, the ping might take indefinite amount of time, and holding the connection permit. We need to add a timeout similar to how we have close timeout.
2. Ping on connection aquire if test_before_acquire is set to true. Unlike return_to_pool, the duration of this is bounded by the acquire timeout. However, if we trigger the acquire connection timeout, the caller will observe an error. However, if we configure a tighter ping timeout, the caller will be able to establish a new connection within the acquire timeout.

To add this, add a configurable and optional ping timeout.

Is this a breaking change?

No. The default ping timeout is None, which means no timeout. I would personally set this to tens of seconds or less in a production environment, but kept the default to None so the change is opt-in.

Testing

Added tests that exercise the success and error (timeout) on acquire and return_to_pool. Duration::ZERO always results in timeout to facilitate making the tests non-flaky and deterministic.