Support multiple parallel slices in ShiftedMetricInterp

[johnomotani]

• ShiftedMetric has supported MYG parallel slices instead of just one yup and one ydown slice, this PR adds similar support to ShiftedMetricInterp
• Also replaces use of BoutMask skip_mask with support for Regions. This is an optimization - I tried one simulation which ran ~10% faster (for the full simulation, I haven't done detailed timing on the toFieldAligned/fromFieldAligned parts)
  • Creates a default region in ZInterpolation to use for interpolation with a non-zero y_offset which replaces the previous creation of skip_mask in ShiftedMetricInterp
  • Allows the user to pass some other region in the ZInterpolation constructor. ZHermiteSpline is not totally optimal if passing a custom region, because the calculation of a z-derivative also requires a region, and at the moment RGN_ALL is used for the z-derivative if a custom region was passed.

[johnomotani]

It wouldn't be hard to split this into 2 PRs if that's preferred (one for 7a6f211 and a second for
1b9754e). The ShiftedMetricInterp changes are built on a slightly changed API for ZInterpolation, so the second PR would depend on the first one.

[ZedThree]

I guess the Region argument is being passed as a pointer so that the default value can be nullptr?

This worries me a little bit, because storing it as unique_ptr means that it actually takes ownership of the passed pointer, which is not obvious from the interface. I think there's at least two ways to fix this.

One would be to take unique_ptr<Region> arguments instead, which makes it much clearer that the interpolation is going to take ownership. See core guidelines R.32.

Another way would be to take it to just take it by value instead. Then there are two ways to do this, the first is to use the default argument as Region<Ind3D> region = {}, and then check region.size() > 0 to see if it was passed. The other is just have two overloads, one with and one without the region argument. The one without can then construct a default value and call the other one.

Storing an empty Region should be pretty lightweight, I think. It will basically have a couple of empty vectors, so total 6 or 8 pointers, plus a couple of ints in Region. unique_ptr has (almost) zero overhead, but we're not making loads of interpolations and passing them around, so not likely to have a massive impact.

A better option might be std::optional, but that's C++17.

[johnomotani]

Thanks @ZedThree, I like the idea of just having a Region member with no pointers! Sorry I forgot to tidy this up before submitting the PR, I'd originally meant to have a unique_ptr argument, but had something wrong (think it might have been a missing & from a const std::string& argument somewhere) and got confused by the compiler error messages with all the templates (unique_ptr argument to the constructor of a class that is created by a GenericFactory subclass...) until I gave up on the unique_ptr argument (but meant to fix it later!).

[ZedThree]

Could store const std::size_t ydown_index = mesh.ystart instead of Mesh& to index parallel_slice_interpolators above? This removes the need to include the mesh header here, although we'd still need in it the .cxx file.

For consistency, might then be nice to have constexpr std::size_t yup_index = 0;

[johnomotani]

The ParallelTransform base class already has a Mesh& mesh member, so I didn't need to add it to ShiftedMetricInterp at all... I think it's still nicer to add yup_index and ydown_index members as you suggested, it's probably nicer to have a const size_t than a Mesh member variable in the hopefully-inlined getWeightsFor* methods.