I had a long post but after catching up on the huge backlog of posts that accrued while I was busy, much has been covered so will make some numbered points instead:
1. I've read this as a 2 stage process - AP off, an initial upset but the crew possibly thinks they have recovered it by 2:10:50 (when one calls for the captain). By this stage, PF has bled off a large amount of speed. Then a second event around 2:10:51 which causes a stall, apparently unrecovered until impact
2. At FL380, assuming actual airspeed was 215kt, and AoA 4 degrees, how far off stall were they as a function of airspeed? Could it be as simple as decaying airspeed at that point, hidden by UAS problems? (With underlying original cause, an unintentional climb, when trying to control roll, that lost them a large amount of airspeed?)
3. If the first two stall warnings were invalid, but the third one (which lasted a minute, and then reappeared briefly when nose was put down) was valid, could it have been disregarded initially because it followed two invalid ones? That is, crying wolf twice, followed by a true warning that was then ignored.
4. Could at least some of the nose-up commands (after the initial left/up) be due to the large divergence between AoA and pitch? If the plane started dropping hard vertically, causing very large AoA, but the ride was relatively smooth (see gums' posts a long while back), and pitch did not appear to be too high, then wouldn't nose-up commands appear logical? ie the pilot sees altitude dropping with nose apparently not too high, does not believe they are in a stall, and tries to lift it? Instead, consolidating the stall? All this even harder to detect at night, inside bad weather, no moon anyway. And with the effect of moving the THS back so far.
5. During the minute that followed the stall, ISIS airspeed recovers. Then after that minute, airspeeds go "invalid" and the third stall warning stops (having been going for a minute). While considered "invalid", was the airspeed either accurately measured <30 kts (or was it low, but further affected by the angle of incidence for pitots) and merely considered to be invalid because it was so low -- and in that case, we have a true (or nearly true) "invalid" airspeed following, and possibly confused by, original false reports caused by pitot problems? (ie again, the low airspeed was initially pitots crying wolf, but the next time it was for real because so much forward airspeed had been lost)
6. A design error that the (third) stall warning deactivated at unreliable airspeeds, only to reactivate once control started being reestablished (nose down, pushing airspeeds up above the "invalid" threshold, and leading to valid computations again) -- but with the net effect of making the correct input appear to be the wrong one.
(Apologies if anything above is self-evidently wrong to a real pilot - happy to be corrected if that's the case)