Fix stored coefficients to make M2L faster for weird kernels (#91)

* Fix stored coefficients to make M2L faster for weird kernels

* remove extra blank lines

* fix formatting

* use mi

* fix laplace

* fix ordering of mis

* split _get_coeff_hyperplanes to a function and override in child class

* rename pde_dict

* rename to axis_permutation

* iterate dict object instead of dict.keys()

* explain mi ordering key

* _get_coeff_hyperplanes -> _get_mi_hyperplanes

* Improve docstrings

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* Improve comment in _get_mi_hyperplanes

Co-authored-by: Andreas Klöckner <inform@tiker.net>

Co-authored-by: Andreas Klöckner <inform@tiker.net>