1000% yes to all points. I also wish there were gpu options in serverless compute and shared attached ssd or other fast storage, not to mention cross function communication.
Nice review. Bit too technical for me.
For persistence, a function can use a disk (shared or not) or a database (in memory or not)
did not get which is used here.
The approach here is that serverless should be planned at the application level. There is another way, make sure the infrastructure handles serverless. All in all, I wish there was examples for this