Once you are under 1µA, the real factor in power consumption is the awake time, not the sleep time…
If you draw 1µA in sleep time, it will require a capacity of 10 mAh over a year, which is nothing compared to the wake-up current draw.
To give an idea I measure 0.84µA on my nodes, with RFM95 and BME280 connected, and I do nothing on the RFM95 to save power (I don’t think you will go under 0.40µA for the 328p alone – but again this is not the challenge, the challenge is to minimize your awake time/consumption)
I am indeed re-considering the power gating option. Initially I was using ABP, therefore the issue never surfaced with the JOIN. But yes, I think using the TPL5010 on a hardware interrupt to wake up the MCU seems the best case scenario.