Add GPU support for extruded 1D spaces

[dennisYatunin]

This PR adds GPU support for extruded 1D spaces, with the following modifications:

• Remove the assumption that Ni == Nj from mapreduce_cuda_kernel, columnwise_partition, columnwise_universal_index, and multiple_field_solve_universal_index
• Add methods for operator_shmem operator_fill_shmem!, and operator_evaluate corresponding to the 1D versions of all SEM operators
• Rename DSSDataTypes to DSSTypes2D, and define a new constant DSSTypes1D
• Add a method for dss_1d! corresponding to CUDADevice, and extend dss! to support both 1D and 2D data types
• Adapt the local_geometry and dss_weights of a SpectralElementSpace1D from Arrays to CuArrays

These changes are also tested through CI:

• Add a CUDA job that runs unit_2d.jl and convergence_2d.jl
• Add tests for IFH, IHF, VIFH, and VIHF to unit_mapreduce.jl, which already has a CUDA job

To fix several test failues, I had to make a few additional changes:

• Simplify the CUDA implementation of Gradient by directly differentiating vectors instead of breaking them up into components, as the original design could not distinguish the gradient of a scalar from the gradient of a single-component vector, and one of the tests in unit_2d.jl takes the gradient of a UVector field
• Extend strip_space to FiniteDifferenceOperator boundary conditions, as one of the tests in unit_2d.jl uses a 1D field for the CurlC2F boundary condition
• Reduce the number of methods for level(space::ExtrudedFiniteDifferenceSpace, v) from 4 to 1 to avoid a GPU inference failure in reconstruct_placeholder_space when evaluating boundary conditions (we hit some sort of compiler heuristic when we use 2 or 4 specialized methods, so we need to just have one method with union-splitting)
  • By making this change, I have inadvertently fixed the issue in gradient on combination of horizontal and 3d fields does not work #1989
• Adapt several CuArrays to Arrays in unit_2d.jl and unit_spaces.jl to avoid scalar indexing errors
• Bump the number of allowed JET failures for SpectralElementSpace1D in opt_spaces.jl from 141 to 825 on GPUs and from 137 to 150 on CPUs

---

• Code follows the style guidelines OR N/A.
• Unit tests are included OR N/A.
• Code is exercised in an integration test OR N/A.
• Documentation has been added/updated OR N/A.