AM modifications + ellipsis selection

[cphyc]

This adds notably

• the AM modification routines
• a finite-difference approach to compute the gradient of the potential (used to compute the Zel'dovich displacement)
• an ellipsis selection routine (useful to make quick AM modifications, as spheres are pathological)

[cphyc]

I am quite confused by the failing tests, and weirdly the tests do not run locally (I get a segfault, including with master).

[cphyc]

@apontzen I realized that this PR is including quite a significant change in the way velocities are computed under the Zeldovich approximation.

The current version of master relies on the ik operator in Fourier space to compute the potential gradient (and hence the velocities). As of 21121769a49cae621c42d91514b846ca222a2d8c, the code is using instead a fourth-order finite-difference scheme. On my machine, this leads to significant differences in the velocities and positions.

For example for the velocity displacement test, I get these differences for the positions and for the velocities

The issue is that the angular momentum computation is only consistent with the output if the Zel'dovich velocities are computed like the AM one, i.e. using the finite-difference scheme.