PJ2 "Once the aircraft 'departed' level flight". Other than level is not a 'departure', yes? Departure means a loss of aerodynamic flight, no?
Per BEA, how is the a/c known to be at S/L flight at dropout? PF obviously corrected for a ND or descent (or overspeed) of some sort, right? Assuming his NU input was a stupid or inadvertent blunder is not supported by the Data, surely? He could have pulled NU to correct an overspeed whilst NU already, yes? This would fit with such an unusual roc so quickly. Would the a/p have trimmed the NU to control speed with added power it had applied to "correct" a "slow" IAS due particle ICE plugging?
I do not mean to nitpick here, but your commentary is the Gold Standard here, and am I missing your drift? Maneuvering to retreat from STALL WARNING is not a STALL recovery, that is why the book allows for back pressure. I cannot find where it allows NU? Only back pressure, to minimize altitude loss? In fact, it advises lowering Pitch at this point, which is not the same as ND. This is parsable, but I read it as "non-negative" 'Target Pitch', to maintain altitude. I find nowhere a direction for PITCH UP.
Situational Awareness is the key, of course. Strictly speaking, loss of autopilot, for whatever reason, is a loss of SA machine-wise? Whether for UAS or inability to maintain programmed flight envelope limits?