Direct loopy kernel (#95)

* post process kernel

* Fix postprocesslocal

* use preprocess/postprocess everywhere

* Fix event management for m2l

* Use a loopy kernel

* return a loopy kernel

* fix formatting

* reduce diff

* Fix typo

* Fix typo

* use_preprocessing_for_m2l -> use_fft_for_m2l

* supports_optimized_m2l -> supports_translation_classes

* restore non fft optimized code path

* events -> timing_events

* More descriptive exception message

* use lp.Assignment instead of strings

* Fix typo

* Add a comment about symbolic sum

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

* Fix error message

* split too long line

* Make use_preprocessing_for_m2l an option

* Add an comment about the options

* fix syntax errors

* add missing defines

* Fix hashing

* Fix returning nexprs

* fix use_preprocessing_for_m2l

* Fix ncoeff_src and ncoeff_tgt

* fix use_preprocessing_for_m2l for bessel based

* Fix logic

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