Adapt pytential's fmmlib support to make use of optimized M2L See merge request inducer/pytential!152
