WIP: clustering, drop partner neurons with type=NA

[jefferis]

• leaving NA types presumably adds some noise; it could be argued that this is a good thing if it leads to lumping cells when a lot of partner neurons cannot be identified
• I anticipate that merging this will change a lot of across dataset clustering slightly
• need to check impact on single dataset clustering (grouped vs ungrouped)
• there should be a pathway to revert to previous behaviour – should think about whether this should be via an option (already an implemented), an explicit function argument or both