fix: Fix targetting class to be neuron compatible

[danilobenozzo]

Describe the work done

• Add an option to get all global IDs with CellTypeFilter when get_targets() method is called.
• Add spherical targetting per cell types.
• ByIdTargetting fixed to use global IDs

[Helveg]

Good addition, just some optimization problems. You create intermediary lists and use N^2 lookup where you probably don't need it

[Helveg]

The ByIdTargetting is inefficient:

• It uses if ids in ps_ids N times, with inside of it ps_ids.index ~> O(2 * n^2). This will explode and never finish if someone copies a list of 1000 cell ids into a 100k cell simulation.
• It's conceptually going roundabout: it relies on ps.load_ids which takes the chunk local ids and combines it with the chunk stats and expands them into global ids, to then look up the position of a global id in the expanded global ids, to then arrive back at the local id.

The fundamental task here is to take an array of global ids and to express it in local ids of a given PS. This is a missing piece of logic it seems (@drodarie could you look through the PS/CS and ConnectivitySetIterator interface classes and bsb-hdf5.. I feel like I wrote this already).

If you want to implement this efficiently, you would have to write a new piece of logic in bsb-hdf5 and/or NeuronPopulation that uses the PlacementSets chunks, and the chunk_stats and uses simple arithmetics to calculate the local id of each global id. @drodarie could you help Danilo piece this together?

[filimarc]

With this commit i will use a method, that i've added to bsb-hdf5, to convert global ids into local ids.
It is linked to: dbbs-lab/bsb-hdf5#40