Diagnostic enhancements of postgkyl

[JunoRavin]

This issue can possible be split into multiple issues/sprints, but for now collecting the most high priority diagnostic enhancements that would streamline scripting analysis of Vlasov and fluids simulations:

• Extensions to primitive moments (and support for Vlasov simulations)
  ** Infrastructure to check if input is "finite-volume-like" or a DG field. If "finite-volume-like" call routines that compute u, T, P_{ij}, etc. (and add the primitives of the third moment)
  ** If data is a DG field, compute the primitive moments using weak operations (see Issue Weak operations in postgkyl #13)

• Drift analysis
  ** Compute total and individual components of parallel and perpendicular flows (see attached image for example of vector perpendicular flow)
  ** By default, could use np.gradient for "finite-volume-like" data and differentiate() for a DG field. Long-term, the accuracy of gradient operations could be improved by the DG field being passed to these operations being a DG field computed from recovery.

• Energization analysis
  ** Using drift analysis, compute total parallel/perpendicular energization, and energization from various components of the flows
  ** Implement additional common energization metrics (J dot E, Zenitani, p-Theta, Pi-D)

• Auxiliary diagnostics
  ** Swisdak measure of agyrotropy

• Field-particle correlation
  ** Total correlation |v|^2 E dot grad_v f
  ** Component-wise correlation v_i E_i df/dv_i
  ** General fixed-time frame-transformed correlation (v_{sim} -> v_{sim} - U, E_{sim} -> E_{sim} + U x B_{sim})
  ** Total parallel correlation ((v dot b_hat) b_hat)^2 * ((E dot b_hat) b_hat) df/d((v dot b_hat) b_hat)
  ** Total perpendicular correlation
  ** General aligned correlation which takes the magnetic field and a flow vector as inputs and computes three new velocity/field variables: v_{parallel} = (v dot b_hat) b_hat, v_{perp_2} = (U x b_hat) dot v (in direction U x b_hat), and v_{perp_1} = (b_hat x (U x b_hat)) dot v (in direction b_hat x (U x b_hat))
  *** Since many of these are total correlations, could have a single function which takes as input the velocity vector, electric field vector, and distribution function, and computes the instantaneous or averaged correlation from these three quantities
  *** Separate functions could be used to construct the transformed variables (and transformed grid to compute derivatives on)