Currently, when building an applyOps, we reserve as many optimes as we have operations to put into the applyOps, and then use the first N (for as many applyOps as we have). This is legal, but it advances the cluster time to the end of the reserved opTimes. This means that if we immediately follow doing an applyOps write with a causal secondary read, we will think the cluster time on the secondary is not up to date.

We should either use the last N slots for the applyOps, or reserve the exact number of slots.