healpix-convolution: convolution on the healpix grid

Convolving on the healpix grid has traditionally been done using spherical harmonics, and thus in the frequency domain.

However, for small kernels this has the non-negligible overhead of transforming from / to spherical harmonics, and thus tends to be slow.

To resolve this, this package instead constructs a sparse 2-dimensional matrix as the kernel, which can then be sparse-multiplied with the data. Additionally, to allow convolving subdomains it also provides tools to pad the data.