Added external force and fixed dtype issues

[hsalehipour]

This PR:

1. adds external force due to exact difference method as new operator
2. adds ForcedCollision as a new way of defining the collision operator (either BGK or KBC)
3. adds a turbulent 3d benchmakr example with reference DNS and theory data for testing 1 and 2 above.
4. fixes a bug in my PR Added boundary force computation and more #61 where interior boundaries that touched the domain bounds would not be properly handled.
5. added a flag to boundary condition to indicate where "index padding" is needed for BCs that required missing directions.
6. fixes all previous pytest issues
7. fixes all issues (caught so far) related to dtype other than fp32 in Warp kernels

[hsalehipour]

Results obtained by running the recently added benchmark example on turbulent channel flow with fp32/fp32. Comparisons are made with DNS data of Moser and Kim (1999) (solid line), as well as the log-law of wall (dashed line)