Allow for permanent grouping of observation

The current grouping algorithm have to be re-run on the whole downloaded observation flat list each time a filter is changed. The new idea is to define groups once for all based on the unfiltered observations.
The grouping algo should be something like all objects from the same instrument/mode a being less than x degrees from another one belong to the same group. The main difference with the previous algo is that when an new observation is inserted and belong to 2 groups, the 2 groups should be merged into 1.
This will allow to have consistent groups, they can even be precomputed on the server side.
Groups can also form a multi-resolution tree, generated by changing distance threshold for each level. This would allow to have e.g. 1 high level group per survey.