xyst

Add partition deactivation to ZalCG

Appears to work, tested with Sedov, starting from an initially deactivated
state dynamically reactivating partitions during time stepping. Gives about 2x
speedup when u=0 compared to u=0.7-0.8 with GreedyRefineLB on N=1 and N=6
compute nodes.

Algorithm ingredients, details:

* Deactivation is based on checking all edges of a chare based on tolerance,
  see ZalCG::dea().

* Reactivation is based on checking chare-boundary edges based on tolerance,
  see ZalCG::rea(), chare-boundary edges: edges of all tetrahedra surrounding
  all tetrahedra with at least one point on the chare-boundary, see
  ZalCG::m_chbndedge.

* Deactivation tolerance can be defined in various ways; currently used is the
  max of the normalized absolute difference of a set of scalar, that can be
  user-specified, e.g., for Sedov: density, see ZalCG::activate().

* Chare node volume contributions must be adjusted for solution, see
  Discretization::m_cvolc.

* Extra mass diffusion is added to all scalar components on edges of the active
  hull to avoid mass unable to leave towards inactive chares, see
  ZalCG::huldif(). The active hull is the edges in ZalCG::m_chbndedge
  interfacing inactive chares.

Next:

* Deactivation is currently only done in the first iteration and ignored
  afterwards.

* Currently reactivation overrides deactivation. This will likely have to be
  modified if/when deactivation is enabled during time stepping.

* A potential way to modify the deactivation/reactivation logic to enable
  deactivation during time stepping is to override
  (1) deactivation with reactivation only if the chare-boundary is outflow, and
  (2) reactivation with deactivation only if the chare-boundary is inflow.