OK, I checked it out more thoroughly this time by using the scope by adjusting the code to set a pin HIGH when SleepNow() was invoked and LOW after it woke up.
Bottom line: It *was* getting prematurely woken up by the PllLock pulse (see previous o-scope pictures above). However, the workaround really does work: just wait until after PllLock goes HIGH, and then sleep in deep powerdown mode. After doing that, nothing else woke it up prematurely. The scope shows that after RSSI goes HIGH, the atmega wakes up 2.1ms later and records the RSSI values at that time.
My initial impression is that it does look as though there may be some reduction in average noise by capturing RSSI while the moteino is asleep, and fortunately it's very easy to do. So, as a practical matter, if listening to very faint signals, it may be worthwhile to put the atmega into deep powerdown sleep while leaving the RFM69 in Rx mode. However, it would probably only matter at the margin, for signals that are very near the noise floor.