Support surface data

[JulioAPeraza]

Closes None.

In this PR, I will start drafting some of the things we need to implement to support surface data in NiMARE.

• Support 2D coordinates of vertices in Kernel transformers: ALEKernel, and KDAKernel.
• Support 2D arrays in compute_kda_ma and compute_ale_ma to compute MA maps on the surface.
• Note: For these cases, we won't need sparse arrays.
• Return 2D kernels from get_ale_kernel when requested.
• Add vertex2voxel and voxel2vertex functions. Use registration fusion (Wu et al., 2018).
• For voxel2vertex (required for surface-based meta-analysis), use a surface-to-MNI mapping (only available for fsaverage 164K, here). This function implemented in Matlab returns the indices of the vertex given the MNI coordinate from volume space.
• For vertex2voxel (not required, only implemented for completeness), use the MNI-to-surface mapping available here. Surfaces available: civet (41k), fsaverage (3k, 10k, 41k, 164k), and fsLR (32k, 164k). See here a Matlab implementation.
• For the kernel transformer, get the geodesic distance between all vertices on the surface. See get_surface_distance.
• Masker:
• In this case, we won't need a masker, but a function that converts gifti to a numpy array. See _gifti_to_array in neuromaps.
• Remove the medial wall from each hemisphere and stack the two arrays: https://github.com/JulioAPeraza/gradec/blob/c2cae635132c2b42b47c36b237fb393eb6ebc158/gradec/utils.py#L121