It's not "fuel" in this case, it's the highly reactive compounds that are used to ignite the engines (TEA/TEB are pyrophoric, so they ignite spontaneously when exposed to air).
The precise amount they'll need depends on how quickly and reliably the engines restart. If the engines take a couple extra shots to light, then you may run out.
No, they have used this profile several times in the past. The past couple droneship landing have used three engines for the landing burn (the outer engines shut down before landing).
The recent "stage that lived" was testing three engines all the way to "0" (which worked fine, apparently).
My guess is the hotter than normal reentry meant the engines were a bit tougher to relight, and they burned through more starter loads than they expected.
I'm not sure what your sequences mean there, almost every boostback and re-entry burn has used three engines (they always light the center engine first, then the outer two).
Most recent droneship landings recently have used three engines for the 'middle' part of the burn, shutting down the other two before actually landing.
The recent GovSat launch tested a three engine burn all the way down (the first time they have tried that, AFAIK).
There was nothing unusual about the number of relights. My guess is the issue was with the difficulty of the relights.
It's not quite that simple. So many factors, as well as random chance, affect the quantity of TEA/TEB used that it's virtually impossible to create an accurate prediction. Other comments in this thread mention that they could probably have modeled and tested on the ground until they had a better idea, but that takes a long time and is expensive. Actually launching the thing and learning from that is a much more effective usage of time and money. As someone else ITT quoted George Box "all models are wrong, but some are useful.".
The precise amount they'll need depends on how quickly and reliably the engines restart. If the engines take a couple extra shots to light, then you may run out.