fuzzy_join: dense vs sparse arrays

[jovan-stojanovic]

@fcharras noticed that using dense arrays rather than sparse for the fuzzy_join significantly increase it's speed. It would be nice to have a benchmark and, if it's the case, change the type of arrays.

[jovan-stojanovic]

#511 to take into account

[fcharras]

@GaelVaroquaux suggests RandomProjection to transform the current array into a dense array. The dimension of the dense array is a parameter to set low enough so that the allocated memory and the speed are acceptable, and we don't know how much the quality of the fuzzy join could suffer from the projection, that should be something to check ? Would there be a thumb rule that give a good enough target dimension with respect to the density of the input array ?