Most of those drives in the endurance tests indeed went read-only, but then bricked themselves after a power reset by design
(I have no idea why anyone would ever consider this the proper way to handle running out of P/E cycles). So the data recovery service would actually need a way to read the chips directly, which generally only the original manufacturer (or NSA because they have unlimited money) can do because only they'll know how everything is organized. In comparison a head swap should be doable by any recovery service
A bad sector can be marked bad by the OS or reallocated to spare sectors by the drive firmware so aren't a big deal. Failed head means the drive locks up when you try to access anything under it. I actually have a few HDD in service with a failed head--I had to partition around the entire bad side of the platter but they still work (the beginning is where formatting crashes and the end is calculated from dividing the total by the number of platter sides). A crashed head is the worst--it can scrape enough material off the platter so the drive sounds like it has sand in it when you shake it, or the bouncing dust quickly sandblasts the other platters so everything is unrecoverable.
It's not just failures--have you ever tried to use an undelete utility on a SSD? Garbage collection seems to be so aggressive that in many cases you only have about ten minutes to use it before it gets zeroed out. Of course after 1759TB
it deletes itself!
I'd also like to remind everyone that MTBF is not expected lifespan: