Hooks for J to talk to libFLANN.

Tested with fresh libFLANN 1.8.4 build and 1.7.1 that 
comes with ubuntu and debian (bonus: it works on the 
Raspberry Pi). 

http://www.cs.ubc.ca/research/flann/

TODO:  
 - consider storing the parameters as well as the tree


What's it for?
Near neighbor calculations on arrays with a modest number of features, d
(aka columns) and a large number of rows, N is O(Nd) when done via
brute force. 
The default "spatial tree" method in libflann allows for more efficient 
search using a hybrid kd-tree (it also supports Locality Hashing, though 
I have yet to test this). Such searches are O(log(N))

Usage:
For multiple queries:

tree=: conew 'jflann'
create__tree dataset
search__tree testset;nn
radsearch__tree moretestset;nn;rad  NB. radius search
paramsof__tree ''                               NB. tells you what the params of the tree are
destroy__tree''

For single queries:
allsearch dataset;testset;nn

Altering from default parameters:
((<vals) setparams <'key1';'key2') create_tree ...
((<vals) setparams <'key1';'key2') allsearch ...