@mm43
That's fair enough, but I don't see how that contradicts what the BEA are saying about UAS being the reason for autopilot disconnect as opposed to turbulence. Even if that is disregarded completely it doesn't support the hypothesis that an autopilot that disconnected between 02:10:04.5 and 02:10:05 ordered a climb that did not begin until 02:10:11 (not that said hypothesis is one of yours).
As I said, it looks to me like the autopilot disconnected in the middle of one of it's turbulence encounter compensation inputs when the nose was on the way back up (i.e. not on the way down as Lyman hypothesises), but the real beginnings of the zoom climb are clearly ordered by the PF's sidestick input.
On autopilot disconnect, the aircraft locks thrust at the last setting (as you say) and holds the pitch where it was being commanded at point of disconnect (in the middle of the correction) allowing time for elevators to return to neutral, which is borne out by the graph settling at 2 degrees nose up for a couple of seconds (just short of the 3.5 degree pitch it was holding before) before the nose-up commands from the sidestick begin to take over and progressively increase pitch attitude to 11.5 degrees nose up.