Use vctrs::new_rcrd to safely store sampling information.

[kylesnap]

ernest_run now stores an ernest_rcrd object, which stores information about each point discarded during a nested sampling run. In addition to making future features easier to develop (e.g., merging runs and parallelization), the size of the ernest_run object is now much smaller, and ernest can readily warn if a run produces results that do not meet the ordered likelihood assumption of the nested sampling integrator. [1] [2]

Other changes include:

• Refactored the core evidence calculation in calculate.ernest_run() to improve clarity and flexibility. Now, if ndraws = 0, the function uses analytical error estimates and normal approximations to simulate log-evidence uncertainty. [1] [2]
• Rewrote the internal logic for calculating log-volumes and log-weights to operate directly on the new ernest_rcrd format, improving reliability and error handling. [1] [2]
• Removed the radial parameter from as_draws and visualize. [1] [2]
• Updated the DESCRIPTION file to add ROpenSci editors as contributors, remove unused dependencies, add new suggested packages (e.g., distributional). [1] [2] [3]

[kylesnap]

False positive on pkgcheck w/ no website; pulling into main.