Revised the way how multiprocessing is called in the 3D transformer (replaced with pool.imap_unordered without initializer). This is as fast as before but has a much smaller memory consumption enabling the algorithm to also run on smaller machines while still highly benefiting from more CPUs. Fixes issue #5

. Updated HISTORY.rst.

Signed-off-by: Daniel Scheffler <danschef@gfz-potsdam.de>