Feature/shift anti lag

[kferjaoui]

• Add single TEM I/O dispatcher and route fast stage moves through it
• Execute preload/overshoot + correction as an atomic sequence (no interleaving with other TEM calls)
• Keep existing UI signal wiring and TEMClient dependency unchanged

[tgruene]

As far as I can tell, this code replaces 'backlash' with 'preload'. The backlash in task_manager.py (line 803) seems to correct for inaccuracies of the instrument, i.e. if 10um are send to pyJEM, the stage would actually only move e.g. 9.9um, and this is corrected by the backlash. This is very different from the lag-correction. in line 786 in task_manager.py sets backlash to 0 in case the stage is already in moving direction, also this seems to lead to undesired behaviour. If the user came the the crystal from the same direction, lag-correction would still be required in order to move back to the crystal after alignment 10um off the crystal, right?

[kferjaoui]

Hi Tim,
I have updated the logic of the lag-corrected moves in the latest commit dda021c. I have tested the routines behavior using a dummy TEM server, so actual testing at the instrument remains necessary.

To implement the “move away -> align -> move back to crystal” workflow, the new code:

1. renames the parameter to preload
2. and in move_parking_with_preload() it always does the two-step sequence via _two_step_sequence()
   (overshoot then step back), regardless of previous direction.

So the old logic that could zero the correction:

# (old idea)
if direction == 0 and last_diff_sign >= 0: preload = 0
...

has been discarded.

Because for this use case it causes the exact undesired behavior you mention: even if you approached from the same direction, you still want the “park away” move to preload the gears so the later Back is repeatable.

Finally, the return is handled by the added back buttons (Back X for translation in the x direction, Back TX for rotation) through the move_back_no_preload() function which sends one straight move by -away (no preload), then clears the stored away value.

Situation: The GUI allows to accumulate multiple fast movements hence the Back button will compensate for the accumulated distance i.e. if you click twice on the button [+10µm], then you will perform 2 lag-corrected moves i.e. +12µm → -2µm then 12µm → -2µm with a final state at +20µm but cogs set in the direction of (-x) and by using the button Back X you'd perform a straight shot of -20µm without backlash. After completion, the Back X button is disabled (as there is nothing to go back to).

N.B: I kept the old routine move_with_backlash() as it is still used in some of the operations implemented by KT such as in subimageMouseClickEvent() or the CenteringTask (stage_centering_task.py). I can update the functionalities to incorporate the logic of the new fct move_parking_with_preload if you are satisfied with the current lag-corrected fast movements.