Update the quicksort to have a nice sequential and threaded version which accept <= operator or lambda function

Remove some warnings, and split P2PR which is P2P only for one over R and P2P which can accept a matrix kernel (which can be one over R or anything else), both approaches support SSE and AVX but have not been tested.

Provided generic functions for the evaluation of the direct interactions (pass matrix kernel as argument).

Add Adaptive Chebyshev sym  kernel (P2M and M2M)