Add double timeout guard to lazy out req

[Raynos]

        if (self.operations) {
            self.operations.checkLastTimeoutTime(now);
            self.operations.popOutReq(self.id);
        }

We had an issue where on all affinity nodes we had about 0.3 timeouts per second to an individual peer for service FOO.

In the eager out_request path we have checkLastTimeoutTime() to kill connections that are timing out.

We need to add this check back to the lazy out request path.

cc @blampe @jcorbin

[Raynos]

Other solutions to consider when a single service peer for service FOO is unhealthy:

• Ship partial affinity, less peers === better least pending ?
• Reduce kValue, less affinity nodes === better local decisions ?
• Penalize timeout error frames in least pending ?
• Bring back per-peer health state machine ?