Change the default `resume_interval` for bridges and connectors to be
the minimum of `health_check_interval` and `request_timeout / 3`.
Also exposes it as a hidden configuration to allow fine tuning.

Before this change, the default values for `resume_interval` meant
that, if a buffer ever got blocked due to resource errors or high
message volumes, then, by the time the buffer would try to resume its
normal operations, almost all requests would have timed out.