[registration] "How to find the reason why the ndt FitnessScore is too large"

Is your feature request related to a problem? Please describe.
We often use NDT for point cloud registration, and it always gives a FitnessScore. If the FitnessScore is small, we are all happy, but when it is large, how can we find the problem?

Context

Especially when the 2 models are about the same, how do we know where is a bad match and where is a good match, instead of giving an average about all voxels.

Expected behavior

Can we visualize different voxel errors, like a depth map to show the difference? This way we can know which part of the point cloud matches well and which part does not match well. Because some nuances are difficult to quantify with the human eye

Current Behavior

Just a FitnessScore and the output point cloud

Describe the solution you'd like

Display the error of each voxel, and sort

addition:
cloudcompare already has this feature and can visualize

Start by locating PCL's NDT registration implementation and the code that computes FitnessScore. Investigate how voxel-level errors could be exposed and visualized, using the issue's CloudCompare reference as context. Done means users can inspect and sort per-voxel errors rather than only receiving an aggregate FitnessScore.