Randomly pivoted Cholesky and positive definite kernels (supersedes PR #85)

[rileyjmurray]

Add initial support for positive definite kernels (see /misc/rl_pdkernels.hh).

• Various helper functions for evaluating the Gaussian kernel (or any kernel that primarily relies on the Euclidean distance matrix).
• Add a block_arrowhead_multiply function to avoid duplicate kernel evaluations when accessing a kernel matrix as a black-box linear operator.
• Add an RBFKernelMatrix class that represents a collection of regularized RBF kernel matrices. This satisfies the SymmetricLinearOperator concept.

add randomly pivoted Cholesky (see /comps/rl_rpchol.hh):

• Utility functions
  • compute_columns evaluates kernel matrices in parallel where the kernel matrix is specified by a function handle callable as K_stateless(i, j).
  • pack_selected_rows
  • downdate_d_and_cdf
• rp_cholesky: implements Algorithm 4 from https://arxiv.org/abs/2304.12465.

Add a new driver, krill_full_rpchol, in drivers/rl_krill.hh, which basically implements arXiv:2304.12465's Algorithm 1.

[rileyjmurray]

@TeachRaccooon this is ready to merge.