Improve robustness of cohort diversity statistics and make cache version easier to manage

[Vibhanshu230]

This PR makes a few small improvements to cohort_diversity_stats() to make the function more robust and easier to maintain.

• Replaced the hardcoded cache version string with a version constant so cache invalidation is easier if the computation logic changes later.
• Added a check to ensure n_jack is smaller than the number of sites to avoid invalid jackknife behaviour.
• Handled a possible divide-by-zero case when computing Tajima’s D if there are no segregating sites.

These changes do not affect the expected outputs in normal cases, but help prevent edge-case errors and make the cache behaviour clearer. I am happy to adjust the implementation if there is a preferred style for validation or cache versioning in this module.