An update, found a really weird symptom. So my old MCU car was connected solid, but the new one starting acting up at nights. So, while it was acting out, I was playing around with various settings like changing local subnet in case it somehow collides with the VPN address space Tesla uses - grasping at straws. Then out of desperation, created a new WiFi network to see whether maybe two Teslas on the same WiFi network are causing an issue and suddenly the new MCU stopped looping, connected to the original WiFi network (not the new one I created) and stayed connected. So, I figured another intermittent Tesla software behavior, so I disabled the new network thinking I will come back to it next time it flakes out, and then the new MCU started having problems connecting again! I re-enabled the new network, new MCU stopped crapping out again (on the original network, not the new one just created/re-enabled)!
So, it seems that Tesla WiFi connection flakeout depends on how many other WiFi networks are discovered at the same time. I am wondering whether Tesla just never tested their WiFi in an urban environment with a shitload of access points available in the area (my access points report "seeing" 129 different wifi networks in the last 24 hrs), and maybe whoever they hired to do the software didn't allocate enough memory to manage that many networks - possibly they corrupt each other and by adding one more my old network went into some different memory slot which was not being corrupted.