[BUG] wrong pose in unimol docking v2

### Describe the bug

ligand pose is wrong or move out the pocket

### Uni-Mol Version

uni-mol docking v2

### Expected behavior

I use internal case to test unimol docking v2.
But I found the so many ligand move out the pocket. It's a terrible mistake. But the pose is beautiful.
And then I try to close the clash fix function, I found the pose is so bad, some molecular pose is like a line. Then I check the code, I found the clash fix function use pose alignment and try to clean the clash by moving the position?
So that's why the molecular move out the pocket?
If it's true, maybe that means unimol docking v2 can't generate the correct pose, all the magic is the post process function.
I use generation conformation by rdkit for input . Did I do something wrong? If not, do you have any plan to update the model and improve the quality of pose?

### To Reproduce

_No response_

### Environment

_No response_

### Additional Context

_No response_

Start by reproducing the reported Uni-Mol docking v2 case with the RDKit-generated input conformation and compare poses with clash fixing enabled and disabled. Inspect the clash-fix function and its pose-alignment step; the issue is resolved only when the source of poses moving out of the pocket is identified and a reproducible correction is defined.

Written by the indexing model from the issue text.