unifying the usage of the segment_anything codebase

[wyli]

Describe the bug
currently there are two different copies of 'segment_anything', would be great to unify the usage and try to avoid direct modifications (use import and inheritance if possible)

1. https://github.com/Project-MONAI/VISTA/tree/main/monailabel/monaivista/lib/model/vista_point_2pt5/segment_anything
2. https://github.com/Project-MONAI/VISTA/tree/main/training/segment_anything

the 2nd version has a good example
https://github.com/Project-MONAI/VISTA/blob/main/training/vista_2pt5d/vista_image_encoder.py#L22