K-S stat different on R2021b leads to different Pareto fit to \tau than other matlab releases

[mgcooper]

Describe the bug
The K-S stat calculation and resulting Pareto distribution fit is different on R2021b than other matlab versions, all else equal, using the bfra_kuparuk.mlx example.

To Reproduce
Steps to reproduce the behavior:

1. Type dbstop in bfra.plfitb at 68 in a matlab command window or open bfra.plfitb and place the dbstop
2. Run the example in demos/bfra_kuparuk.mlx on Matlab R2021b and any other of the following releases: R2020b, R2022b
3. When the code enters debug mode on line 68, evaluate line 68 and plot the K-S stat (variable D)
4. Repeat in R2020b
5. Alternatively, step into the function bfra.deps.plfit when in debug mode on line 68 of bfra.plfitb and run the code that computes the k-s stat and plot the result.

Expected behavior
The K-S stat should be identical leading to identical Pareto distribution fits to the example data in bfra_kuparuk.mlx

Screenshots
R2021b (y-axis is K-S stat):

R2020b (same as R2022b):

Desktop (please complete the following information):

• OS: macos
• Version 12.6.3

Additional context
Mainly relevant for replicating the results in bfra_kuparuk.mlx (and Cooper et al. 2023) but if you are conducting analysis in R2021b, results may be affected by this.