Improve bilevel particle advection
Opencl advection kernel modification for bilevel high order interpolation (commits 0e7efc09 and 78460030) was not fully correct and performant. These commits was inverted in the mpi-operator branch by commit 3a2d689d
TODO :
- Implement a not full line cache for velocity bilinear interpolation (bilevel)
- Implement a higher order interpolation for velocity in a separate operator
- Add support for resolution larger than max dimension of a thread block.