Small update if some are interested. I've got the new board built with a voltage supervisor and it works well. Once programmed, you simply plug it in and it powers up. Turn on of the ESP happens around 3.25V and it disables around 2.65V. I added a couple lines of code to wait for the vcc to reach 3.3V before booting up.
At first I was confused by how much current the TSS521 consumes (~1.2ma on my inital build), then I read into the HAN protocol. Slaves are supposed to consume a fixed amount of current as they modulate their current to communicate with the master. Obviously with the buck on the input this design does not conform. Input currents on my design range from 0.5ma (TSS521 only, idle buck), to 7.5ma (when bucking from 24 to 3.3), and 15ma (when bucking from 12 to 3.3). This said, I'm not sure how critical the constant current draw is for this application. I've had such a system running on my han for the last several months without issue. However, I guess they could update the FW on the meter and it could develop problems. I'm slowly working on a method for solving this (current sense feedback into a amplifer to burn the extra current). If others have clever ideas here I'm all ears.
This said it runs okay on my Aidon meter with 2.5second messages. I'm averaging ~5.5ma of current consumption from the HAN port. With the slower message speeds of the Kamstrup I think it will consume less. I'll need to put a voltage divider on the run pin of the buck controller so it only pulls power when at the 24V level. Is there still any interest of others for such a device?