Ensure pixel SOM training reruns if new markers specified

[alex-l-kong]

What is the purpose of this PR?

Closes #887. It's currently not possible to re-run 2_Pixie_Cluster_Pixels in case the user needs to specify a new subset of markers on the same cohort. This is a common occurrence and should be addressed.

How did you implement your changes

In PixelSOMCluster.train_som, add a check to verify the existing weights columns (if applicable) and the provided set of columns match. If they do not, this means a new set of columns was specified and PixieSOMCluster.train_som should re-run.

[cliu72]

I think it would be a good idea to add this functionality to train_som in the cell version as well. I think it could come up in the general cell clustering case (for example, in Lorenz's case if he wants to re-run the training and leave out some markers)

[alex-l-kong]

I think it would be a good idea to add this functionality to train_som in the cell version as well. I think it could come up in the general cell clustering case (for example, in Lorenz's case if he wants to re-run the training and leave out some markers)

Yeah that's a good point, I had mentioned it in the discussion on the generalized cell clustering PR but definitely better to do it here.

[cliu72]

Looks good!