Accelerated partitioning

[JulienPeloton]

This is an aborted PR in which I tried to accelerate the partitioning by filtering the dataset prior to repartitioning:

// Load data
df = spark....load()

// Add a column with future repartitioning
df_colid = df.prePartition(...)

// Keep only data of interest
df_colid_subset = df_colid.filter(cond)

// Repartition only data of interest
df_repart = df_colid_subset.repartitionByCol(...)

But it turns out to be super slow, because of the udf in prePartition.

So I decided to delay this development, but still merging other minor changes (docs + getNeighborNodes) which are useful.

TBC.